Motivations for Speeding
Volume II: Findings Report DISCLAIMER
This publication is distributed by the U.S. Department of Transportation, National Highway and conclusions expressed in this publication are those of the authors and not necessarily those
The United States Government assumes no liability for its contents or use thereof. If trade names, to the object of the publication and should not be construed as an endorsement. The United States Government does not endorse products or manufacturers.
Suggested APA Format Citation:
Richard, C. M., Campbell, J. L., Lichty, M. G., Brown, J. L., Chrysler, S., Lee, J. D., Boyle, L., & Reagle, G. (2013, September) Motivations for Speeding, Volume II: Findings Report. (Report No.
This report is accompanied by:
Richard, C. M., Campbell, J. L., Lichty, M. G., Brown, J. L., Chrysler, S., Lee, J. D., Boyle, L., & Reagle, G. (2012, August). Motivations for Speeding, Volume I: Summary Report. (Report
Richard, C. M., Campbell, J. L., Lichty, M. G., Brown, J. L., Chrysler, S., Lee, J. D., Boyle, L., & Reagle, G. (2013, August). Motivations for Speeding, Volume III: Appendices. (Report No. Technical Report Documentation Page 1. Report No. 2. Government Accession No. 3. Recipient’s Catalog No. DOT HS 811 818 4. Title and Subtitle 5. Report Date Motivations for Speeding, Volume II: Findings Report September 2013 6. Performing Organization Code
7. Author(s) 8. Performing Organization Report No. Christian M. Richard, John L. Campbell, Monica G. Lichty, James L. Brown, Susan Chrysler (Texas Transportation Institute), John D. Lee (University of Wisconsin, Madison), Linda Boyle (University of Washington), George Reagle (George Reagle and Associates) 9. Performing Organization Name and Address 10. Work Unit No. (TRAIS) Battelle Memorial Institute 505 King Avenue 11. Contract or Grant No. Columbus, Ohio 43201-2696 DTNH22-06-D-00040, Task#2 12. Sponsoring Agency Name and Address 13. Type of Report and Period Covered National Highway Traffic Safety Administration FINAL REPORT Office of Behavioral Safety Research August 2008 – December 2011 1200 New Jersey Avenue SE. 14. Sponsoring Agency Code Washington, DC 20590 15. Supplementary Notes Geoffrey Collier (initial) and Randolph Atkins (final) were the Contracting Officer’s Technical Representatives (COTRs). 16. Abstract
This is Volume II of a three-volume report. It contains the results of a study that examined the speeding behavior of drivers in their own vehicles over the course of three to four weeks of naturalistic driving in urban (Seattle, WA) and rural (College Station, TX) settings. The purpose of this research was to (1) identify the reasons why drivers speed, (2) model the relative roles of situational, demographic, and personality factors in predicting travel speeds, (3) classify speeders, and (4) identify interventions/countermeasures and strategies for reducing speeding behaviors. Data collected from 164 drivers included 1-Hz recordings of vehicle position and speed using GPS receivers, responses to a battery of a personal inventory questionnaires, and daily driving logs that captured trip-specific situational factors. Vehicle speed and position data were combined with road network data containing validated posted speed information to identify speeding episodes. The descriptive analysis of speeding data provided evidence for different types of speeding behaviors among individual drivers including: (1) infrequent or incidental speeding, which may be unintentional, (2) trip-specific situational speeding, (3) taking many trips with a small amount of speeding per trip (i.e., casual speeding), and (4) habitual or chronic speeding. Regression models were developed to identify predictors of “any” speeding (logistic regression) and amount of speeding (linear regressions). Significant predictors included demographic variables such as age and gender, situational factors such as time-of-day and day-of-week, and key personal inventory factors such as attitudes towards reckless driving. In addition, focus group discussions were conducted with a subset of study participants who were classified as “speeders” and “non-speeders” to identify key attitudes and beliefs towards speeding and the effectiveness of potential countermeasures. 17. Key Words 18. Distribution Statement Speeding, Speed Selection, Free-flow Driving, Speeding Document is available to the Countermeasures, Unsafe Driving public from the National Technical Information Service www.ntis.gov 19 Security Classif. (of this report) 20. Security Classif. (of this page) 21 No. of Pages 22. Price Unclassified Unclassified 204 Form DOT F 1700.7 (8/72) Reproduction of completed page authorized
i
Acknowledgements
The Motivations for Speeding project involved a collaborative effort among numerous contributors, and only the key researchers are formally included as Final Report authors. We specifically acknowledge the efforts and contributions of Diane Williams, Mark Tianow, Ashley Loving, Elizabeth Jackson, Ta Liu, and Dale Rhoda of Battelle, and of the following researchers at Texas Transportation Institute, Katie Connell, Laura Higgins, and Marshall Ward.
ii
Table of Contents Chapter 1: Background and Introduction ...... 1 Objectives ...... 2 Overview of Technical Approach ...... 2 Chapter 2: Methods ...... 5 Introduction ...... 5 Experimental Design ...... 5 Dependent Variables and Covariates ...... 6 Participants ...... 7 Participant Recruitment ...... 8 Participant Incentive Structure ...... 8 Apparatus and Test Instruments ...... 9 Pre-Drive and Post-Drive Personal Inventory ...... 9 Trip Log ...... 10 Device Selection and Pilot Testing ...... 12 Device Preparation and Installation ...... 13 Device Installation and Removal Procedures ...... 16 Data Collection Procedures ...... 18 Enrollment and Start-up Activities ...... 18 On-Road Data Collection ...... 19 Close-Out and Debriefing Activities ...... 19 Data Processing Activities ...... 19 Data Validation Activities ...... 21 Final Data Filtering ...... 21 Definition of Speeding ...... 25 Chapter 3: On-Road Data Collection Results ...... 32 Introduction ...... 32 Descriptive Statistics ...... 32 Overview ...... 32 Seattle Data ...... 35 Seattle Data – All Driving ...... 43 Seattle Data – Free-Flow Driving ...... 50 Seattle Data – Speeding Time ...... 54 Texas Data ...... 60 Texas Data – All Driving ...... 71 Texas Data – Free-Flow Driving ...... 79 Texas Data – Speeding Time ...... 83 Types of Speeding – Seattle and Texas ...... 87
iii
Inferential Statistics ...... 93 Logistic Regression Models ...... 97 Logistic Regression with Personal Inventory Factors and Socioeconomic Variables ...... 104 Linear Regression Models ...... 113 Summary of Regression Modeling ...... 126 Chapter 4: Focus Groups ...... 132 Introduction ...... 132 Methods ...... 132 Participants ...... 132 Transcription Abbreviations ...... 133 Focus Group Activities ...... 134 Data Analysis ...... 134 Results ...... 135 Self-Reported Travel Speeds ...... 135 Interpretation of Posted Speed ...... 136 Perception of Ticket Speed ...... 137 Perception of Safe Speed ...... 138 Factors that Increase Driver Travel Speed ...... 139 Factors that Decrease Driver Travel Speed ...... 141 Countermeasure Discussions ...... 146 Summary of Countermeasure Discussions ...... 163 Chapter 5: Conclusions ...... 165 General Conclusions ...... 165 What is the relative importance of situational factors, demographics, and personality in predicting episodes of speeding? ...... 165 What are the subtypes of drivers with respect to speeding? ...... 170 To what extent are these classes of speeders defined by demographics? ...... 171 To what extent is the predilection to speed correlated with the tendency towards other unsafe driving acts? ...... 171 What attitudes, habits, and behaviors are directly or indirectly related to possible countermeasures, so that subgroups can be compared on these measures? ...... 172 A comprehensive framework of speeding is needed to guide countermeasure development...... 173 Methodological Conclusions ...... 175 How useful was the overall approach methodology in this study? ...... 175 How useful was our definition of speeding? ...... 175 Chapter 6: Methodological Considerations ...... 178 Absolute Estimates of Speeding Time ...... 178 Estimate of “Opportunity to Speed” ...... 178 Time/Duration Criterion of Speeding ...... 178 Excluded Driving Data in Texas...... 179
iv
Generalizability of Personal Inventory Items ...... 179 References ...... 181
v
List of Figures
Figure 1. Overview of proposed approach for studying speeding in drivers and identifying countermeasures...... 3 Figure 2. Daily trip log data fields for a single trip...... 11 Figure 3. GeoChron GPS data logger prior to modification...... 13 Figure 4. GPS installation using the windshield mount option...... 15 Figure 5. GPS installation using the dash mount option...... 15 Figure 6. GPS installations using exterior and interior antenna mount options...... 16 Figure 7. Overview of participant activities...... 18 Figure 8. Data processing flowchart...... 20 Figure 9. Theoretical speed ranges...... 28 Figure 10. Histogram showing household income for Seattle drivers...... 36 Figure 11. Histogram showing highest level of education obtained for Seattle drivers...... 36 Figure 12. Histogram showing marital status for Seattle drivers...... 37 Figure 13. Histogram showing status with regard to children for Seattle drivers...... 37 Figure 14. Histogram showing self-reported weekly mileage for Seattle drivers...... 38 Figure 15. Results of general travel questions for Seattle drivers...... 39 Figure 16. Results of driver behavior questions for Seattle drivers...... 40 Figure 17. Results of risk questions for Seattle drivers...... 41 Figure 18. Results of speeding beliefs, attitudes, and social norms questions for Seattle drivers...... 42
Figure 19. Seattle – Sensation Seeking (Zss) and Impulsivity (Zimp) subscale scores by demographic group...... 43 Figure 20. Map showing the geographic extent of all driving epochs in Seattle...... 44 Figure 21. Histogram showing the total number of trips taken on each day of the week for all Seattle drivers combined...... 47 Figure 22. Histogram showing the average number of trips taken on each day of the week for each demographic group in Seattle...... 48 Figure 23. Histogram showing the total number of miles driven by all participants on Seattle- area roads with different posted speeds (in mph). The colored triangles denote the posted speeds selected for analysis in the 30-35 mph speed band (purple) and the 55- 60 mph speed band (green)...... 49 Figure 24. Histogram showing the average number of miles driven on 30-35 and 55-60 mph roads in Seattle by demographic group...... 49 Figure 25. Map showing the geographic extent of the Seattle free-flow epochs...... 53 Figure 26. Percentage of trips with free-flow time that have speeding time in Seattle, by demographic group...... 55 Figure 27. Scatter plot of percentage trips in Seattle with any speeding in each speed band for individual drivers. The trend line is based on drivers from all demographic groups...... 56
vi
Figure 28. Illustration of how driving data are extracted from a trip and categorized into free- flow and speeding time...... 58 Figure 29. Scatter plot of average - speeding in Seattle (only for trips with any speeding) in each speed band for individual drivers. The trend line is based on drivers from all demographic groups...... 60 Figure 30. Histogram showing household income for Texas drivers...... 61 Figure 31. Histogram showing highest level of education obtained for Texas drivers...... 62 Figure 32. Histogram showing marital status for Texas drivers...... 62 Figure 33. Histogram showing status with regard to children for Texas drivers...... 63 Figure 34. Histogram showing self-reported weekly mileage for Texas drivers...... 63 Figure 35. Results of general travel questions for Texas drivers...... 64 Figure 36. Results of driver behavior questions for Texas drivers...... 65 Figure 37. Results of risk questions for Texas drivers...... 66 Figure 38. Results of speeding beliefs, attitudes, and social norms questions for Texas drivers...... 67 Figure 39. Texas – Sensation Seeking (Zss) and/Impulsivity (Zimp) subscale scores by demographic group...... 68 Figure 40. Map showing the geographic extent of all driving epochs in the Texas data- collection area...... 72 Figure 41. Histogram showing the total number of trips taken on each day of the week for all Texas drivers combined...... 75 Figure 42. Histogram showing the average number of trips taken on each day of the week for each demographic group in Texas...... 76 Figure 43. Histogram showing the total number of miles driven by all participants on Texas roads with different posted speeds (in mph). The colored triangles denote the posted speeds selected for analysis in the 30-35 mph speed band (purple), the 55-60 mph speed band (green), and the 70 mph speed band (yellow)...... 77 Figure 44. Histogram showing the average number of miles driven on 30-35, 55-60, and 70 mph roads in Texas by demographic group...... 78 Figure 45. Map showing the geographic extent of the Texas free-flow epochs...... 81 Figure 46. Percentage of trips with free-flow time that have speeding time in Texas, by demographic group...... 84 Figure 47. Scatter plot of percentage trips in Texas with any speeding in the 30-35 mph and 55-60 mph speed bands for individual drivers. The trend line is based on drivers from all demographic groups...... 85 Figure 48. Scatter plot of average speeding in Texas (only for trips with any speeding) in each speed band for individual drivers. The trend line is based on drivers from all demographic groups...... 86 Figure 49. Types of speeding based on the scatter plot of the primary dependent measures...... 88 Figure 50. Seattle – Scatter plots of types of speeders in each zone...... 89 Figure 51. Texas – Scatter plots of types of speeders in each zone...... 92 Figure 52. Increased penalties and enforcement handout...... 147 Figure 53. Roadway countermeasures handout...... 150 Figure 54. Vehicle-based countermeasures handout...... 153
vii
Figure 55. Automatic enforcement handout...... 157 Figure 56. Speed awareness course handout...... 160 Figure 57. Conceptual framework showing the relationship between driver, vehicle, roadway, and environment factors and corresponding driver performance/profiles...... 174
viii
List of Tables
Table 1. Factors found to be associated with speeding in previous research...... 1 Table 2. Study design and allocation of participants...... 6 Table 3. Data used in the analyses...... 6 Table 4. Driving-related variables recorded for each 30-second driving epoch/record...... 7 Table 5. Distribution of participant demographics...... 8 Table 6. Questions included in the personal inventory...... 10 Table 7. Seattle data filtering results...... 23 Table 8. Texas data filtering results...... 24 Table 9. Different definitions of speeding...... 27 Table 10. Speed behavior types...... 29 Table 11. Final filtering thresholds used to define speeding-event epochs. The right side of the figure shows the number of 30-second epochs and seconds (in parentheses) retained. The Type 3 and Type 4 combined data were used in the analyses to represent speeding...... 31 Table 12. Definitions of key terms used to calculate speeding measures...... 33 Table 13. Number of trips by duration and start time for Seattle drivers...... 45 Table 14. Average number of trips by duration and start time for each demographic group in Seattle...... 46 Table 15. Key trip-level characteristics for drivers in Seattle...... 50 Table 16. Comparison of the number of Seattle driving epochs in each demographic category before and after filtering to retain only free-flow driving epochs on 30-35 and 55- 60 mph roads...... 51 Table 17. Percentage of free-flow driving time in Seattle spent in different speed bands relative to the posted speed (PS) for each demographic group on 30-35 and 55- 60 mph roads...... 54 Table 18. Number of trips with free-flow driving and the percentage of the trip that was speeding by demographic group in Seattle...... 56 Table 19. Average speeding for all trips with free-flow time and only trips with speeding by speed band in Seattle...... 59 Table 20. Speeding beliefs, attitudes, and social norms in Texas and Seattle. Questions that were significantly higher are marked in red for Seattle, and blue for Texas...... 70 Table 21. Number of trips by duration and start time for Texas drivers...... 73 Table 22. Average number of trips by duration and start time for each demographic group in Texas...... 74 Table 23. Key trip-level characteristics for drivers in Texas...... 78 Table 24. Comparison of the number of Texas driving epochs in each demographic category before and after filtering to retain only free-flow driving epochs on 30-35, 55-60, and 70 mph roads...... 79
ix
Table 25. Percentage of free-flow driving time in Texas spent in different speed bands relative to the posted speed for each demographic group on 30-35, 55-60, and 70 mph roads...... 83 Table 26. Number of trips with free-flow driving and the percentage of the trip that was speeding by demographic group in Texas...... 84 Table 27. Average speeding for all trips with free-flow time and only trips with speeding by speed band in Texas...... 86 Table 28. Variables and data used in the regression models...... 95 Table 29. Description of trip variables...... 96 Table 30. Numbers of drivers that had any speeding or no speeding in each speed band...... 97 Table 31. Output from the logistic regression analysis showing the relationship between demographic and trip variables: Seattle 30-35 mph roadways...... 99 Table 32. Comparison of odds ratios across demographic groups...... 99 Table 33. Output from the logistic regression analysis showing the relationship between demographic and trip variables: Seattle 55-60 mph roadways...... 100 Table 34. Comparison of odds ratios across demographic groups...... 101 Table 35. Output from the logistic regression analysis showing the relationship between demographic and trip variables: Texas 30-35 mph roadways...... 102 Table 36. Comparison of odds ratios across demographic groups...... 102 Table 37. Output from the logistic regression analysis showing the relationship between demographic and trip variables: Texas 55-60 mph roadways...... 103 Table 38. Comparison of odds ratios across demographic groups...... 103 Table 39. Output from the logistic regression analysis showing the relationship between demographic and trip variables: Texas 70 mph roadways...... 104 Table 40. Comparison of odds ratios across demographic groups...... 104 Table 41. Names and descriptions of factors obtained in the factor analysis...... 105 Table 42. Output from the logistic regression analysis showing the relationship between demographic, trip, and factor-score variables: Seattle 30-35 mph roadways...... 106 Table 43. Output from the logistic regression analysis showing the relationship between demographic, trip, and factor-score variables: Seattle 55-60 mph roadways...... 108 Table 44. Output from the logistic regression analysis showing the relationship between demographic, trip, and factor-score variables: Texas 30-35 mph roadways...... 109 Table 45. Output from the logistic regression analysis showing the relationship between demographic, trip, and factor-score variables: Texas 55-60 mph roadways...... 110 Table 46. Output from the logistic regression analysis showing the relationship between demographic, trip, and factor-score variables: Texas 70 mph roadways...... 112 Table 47. Numbers of drivers in each demographic group with speeding in each speed band...... 113 Table 48. Graphs showing the transformations of the dependent variable: percentage trip speeding...... 115 Table 49. Output from the linear regression analysis showing the relationship between demographic and trip variables and the proportion of free-flow driving that is speeding: Seattle 30-55 mph roadways...... 116
x
Table 50. Output from the linear regression analysis showing the relationship between demographic and trip variables and the proportion of free-flow driving that is speeding: Seattle 55-60 mph roadways...... 117 Table 51. Output from the linear regression analysis showing the relationship between demographic and trip variables and the proportion of free-flow driving that is speeding: Texas 30-35 mph roadways...... 118 Table 52. Output from the linear regression analysis showing the relationship between demographic and trip variables and the proportion of free-flow driving that is speeding: Texas 55-60 mph roadways...... 119 Table 53. Output from the linear regression analysis showing the relationship between demographic and trip variables and the proportion of free-flow driving that is speeding: Texas 70 mph roadways...... 120 Table 54. Output from the linear regression analysis showing the relationship between demographic, socioeconomic, trip, and factor-score variables and the proportion of free-flow driving that is speeding: Seattle 30-55 mph roadways...... 121 Table 55. Output from the linear regression analysis showing the relationship between demographic, socioeconomic, trip, and factor-score variables and the proportion of free-flow driving that is speeding: Seattle 55-60 mph roadways...... 122 Table 56. Output from the linear regression analysis showing the relationship between demographic, socioeconomic, trip, and factor-score variables and the proportion of free-flow driving that is speeding: Texas 30-55 mph roadways...... 123 Table 57. Output from the linear regression analysis showing the relationship between demographic, socioeconomic, trip, and factor-score variables and the proportion of free-flow driving that is speeding: Texas 55-60 mph roadways...... 124 Table 58. Output from the linear regression analysis showing the relationship between demographic, socioeconomic, trip, and factor-score variables and the proportion of free-flow driving that is speeding: Texas 70 mph roadways...... 125 Table 59. Summary of the five logistic regression models conducted with trip and demographic variables...... 127 Table 60. Summary of the five logistic regression models conducted with trip, demographic, and factor-score variables...... 128 Table 61. Summary of the five linear regression models conducted with trip and demographic variables...... 129 Table 62. Summary of the five linear regression models conducted with trip, demographic, and factor-score variables...... 130 Table 63. Demographic composition of each focus group...... 133 Table 64. Summary of driver responses to the countermeasures presented in the focus groups...... 163 Table 65. Summary of demographic differences based on the likelihood of any speeding on a trip...... 167 Table 66. Summary of factor-score variables that were significant predictors of speeding behavior...... 169 Table 67. List of models in which factors related to unsafe acts were predictors of speeding...... 171
xi
Abbreviations and Acronyms
BB-Safety ...... Behavioral Beliefs related to Safety BCB-Temptation ...... Behavioral and Control beliefs related to Temptation to Speed CB-Opportunity ...... Control Beliefs related to Opportunity to Speed CFCC ...... Census Feature Class Code DBQ ...... Driver Behavior Questionnaire DVRE ...... Driver, Vehicle, Roadway, and Environment FAA ...... Federal Aviation Administration FM ...... Farm to Market GB ...... Gigabyte GED ...... General Education/Equivalency Diploma GIS ...... Geographic Information System GPS ...... Global Positioning System HOV ...... High Occupancy Vehicle IRB ...... Institutional Review Board NCHRP ...... National Cooperative Highway Research Program NHTSA ...... National Highway Traffic Safety Administration NMEA ...... National Marine Electronics Association OR ...... Odds Ratio PS ...... Posted Speed RMC ...... Recommended Minimum Communication SD ...... Secure Digital SOFN ...... Seattle Older Female Non-speeder SOFS ...... Seattle Older Female Speeder SOMN ...... Seattle Older Male Non-speeder SOMS ...... Seattle Older Male Speeder SUV ...... Sport Utility Vehicle SYFN ...... Seattle Younger Female Non-speeder SYFS ...... Seattle Younger Female Speeder SYMN ...... Seattle Younger Male Non-speeder SYMS ...... Seattle Younger Male Speeder ToD ...... Time of Day TOFN ...... Texas Older Female Non-speeder
xii
TOFS ...... Texas Older Female Speeder TOMN ...... Texas Older Male Non-speeder TOMS ...... Texas Older Male Speeder TPB ...... Theory of Planned Behavior TXDOT ...... Texas Department of Transportation TYFN ...... Texas Younger Female Non-speeder TYFS ...... Texas Younger Female Speeder TYMN ...... Texas Younger Male Non-speeder TYMS ...... Texas Younger Male Speeder UTC ...... Coordinated Universal Time WAAS ...... Wide Area Augmentation System WSDOT ...... Washington State Department of Transportation ZKPQ ...... Zuckerman-Kuhlman Personality Questionnaire
xiii
Chapter 1: Background and Introduction
Although speeding is one of the most significant contributors to crash severity and traffic fatalities, attempts to address this problem through a variety of approaches have not led to a significant reduction in speed-related fatalities (National Highway Traffic Safety Administration (NHTSA), 2005). NHTSA’s (2007) report, Countermeasures That Work, provides a list of speeding countermeasures that have been demonstrated to be effective; most of which focus on enforcement or punishment to reduce speeding. A limitation with these types of countermeasures is that they may not be as effective with some driver groups, such as risk-taking young males, who rarely consider the potential consequences of their behaviors (McKenna & Horswill, 2006). Similarly, National Cooperative Highway Research Program (NCHRP) Report 500 provides several engineering-based countermeasures to address speeding (NCHRP, 2005). These countermeasures can be effective in reducing speeding at specific locations, but they can be expensive and only cover small parts of the transportation network, and other non-speeding drivers are also impacted by these countermeasures. Thus, there is a strong interest in new countermeasures to better target identifiable groups of speeders using approaches that specifically and effectively address the reasons for their speeding.
A substantial amount of research has been conducted on the causes of speeding. Table 1 shows the multitude of factors that have been found to be associated with speeding or speed-related crashes. Despite all this research, there is still uncertainty regarding the relative importance of these factors, and how this information can be used to develop countermeasures that effectively target specific types of drivers. This project provides an opportunity to look at several of these factors in a single study and, importantly, to match findings about specific factors to on-road speeding behavior.
Table 1. Factors found to be associated with speeding in previous research. Factor Example Variables Example References Demographic Age, Gender, Socio- DePelsmacker & Janssens, 2006; Harré et al., 1996; Hemenway & Solnick, economic & Education Level 1993; Stradling et al., 2002 Personality Attitudes, Habits, Personal & Arnett et al., 1997; Clément & Jonah, 1984; DePelsmacker & Janssens, Social Norms, Thrill-seeking, 2006; Ekos Research Associates, 2007; Gabany et al., 1997; McKenna & Beliefs Horswill, 2006; Stradling et al., 2002 Roadway Posted Speed Book & Smigielski, 1999; Giles, 2004 Environment Urban/Rural Giles, 2004; Rakauskas et al., 2007 Vehicle Engine Size; Vehicle Age Hirsh, 1986; Stradling et al., 2002 Risky Drinking & Driving, Seatbelt Arnett et al., 1997; Cooper, 1997; Gabany et al., 1997; Harré et al., 1996; Behaviors Use, Red-light Running Hemenway & Solnick, 1993; Rajalin, 1994 Situational Trip time, Mood, Inattention, Arnett et al., 1997; Ekos Research Associates, 2007; Gabany et al., 1997; Fatigue Hirsh, 1986; McKenna, 2005, McKenna & Horswill, 2006
1
Objectives
The purposes of this research effort are to: Identify the reasons why drivers speed, Model the relative roles of situational, demographic, and personality factors in predicting travel speeds, Classify speeders, and Identify interventions/countermeasures and strategies for reducing speeding behaviors.
Specific questions that the project sought to address include: 1. What is the relative importance of situational factors, demographics, and personality in predicting episodes of speeding? 2. What are the subtypes of drivers with respect to speeding? Can we replicate (or fail to replicate) and refine the classification of speeders into two groups: pragmatic speeders and rebels? 3. To what extent are these classes of speeders defined by demographics? To what extent are speeders, in general, defined by demographics? Specifically, are the subtypes of speeding drivers the same across demographic groups? 4. For the sample studied, to what extent is the predilection to speed correlated with the tendency towards other unsafe driving acts? 5. What attitudes, habits, and behaviors are directly or indirectly related to possible countermeasures (points systems, points reduction classes, checkpoints, automated enforcement, crackdowns, traffic calming measures, etc.), so that subgroups can be compared on these measures?
In short, this effort focused on identifying “who” is speeding, “what” contributes to their decisions to speed, and it attempted to identify “which” countermeasures can be most effectively focused at those drivers who account for a significant proportion of the speeding problem. The research consisted of two phases. In Phase 1, the Battelle Team has examined individual drivers’ travel speeds in an on-road study; in Phase 2, we conducted focus groups of select speeder subgroups.
Overview of Technical Approach
The research involved a comprehensive approach for studying speeding in drivers and identifying appropriate countermeasures (see Figure 1 below). The field data collection portion drew upon existing research to identify driver and situational factors associated with speeding, and used this information to develop data collection tools, including a personal inventory questionnaire, a Global Positioning System (GPS)-based, on-road speed recorder, and daily driving-logs that capture situational factors. The information provided by these tools was used to address the objectives listed above, including questions related to: (1) the relative importance of
2
situational, demographic and personality factors; (2) the classification of speeder subtypes; (3) the predictiveness of demographics for speeder subtypes; and (4) correlations between speeding and unsafe driving acts. Following Phase 1 (the field data collection), information about speeder subtypes was used to conduct focus group studies on a subset of participants that match speeder subtypes. This qualitative information addresses the broader project goal of obtaining driver attitudes and behaviors related to countermeasures.
Conduct Plan Study Conduct Field Study Focus Groups
Identify Countermeasures 1 GIS Data Relative Focus Groups Importance of •Road Class Countermeasure Situational, Equipment •Posted Speed Options Requirements Demographics and Personality On-road Data 5 Driver Views on Lit Define Driving Speed/Driving Data 2 Countermeasures Review Measures Classification of Speeder Subtypes Daily Logs Identify Driver Situational Data 3 Factors Predictiveness of Demographics on Survey Speeder Subtype Driver, 4 Demographic, Correlation Urban/Rural between Speeding Characteristics and Unsafe Driving Acts = Research Questions Risky Behaviors
Figure 1. Overview of proposed approach for studying speeding in drivers and identifying countermeasures.
This report provides the methods, results, and conclusions associated with the field data and focus group data collection. Specifically, this report describes a naturalistic field study in which 164 volunteer drivers from Seattle, Washington, and College Station, Texas, agreed to have GPS units installed in their vehicles for 3 to 4 weeks. The GPS data obtained in the study was used to compare the drivers’ speeds to the posted speed limits, referred to in this report as posted speeds (PSs), associated with the roads that they were driving on at any given point in time. These comparisons between actual speeds and posted speeds formed the basis for the analyses and findings described in this report.
This report consists of a number of chapters and appendices, as described below: Chapter 2 describes the methods associated with the on-road data collection effort, including descriptions of the experimental design, study participants, recruiting procedures, driver
3
questionnaires, GPS devices used in the study, detailed data collection procedures, and procedures for processing, validating, and filtering the resulting on-road data. Chapter 3 describes the initial results from the field study, including the findings from a series of analyses providing basic descriptive information about the data, as well as the results of inferential analyses focused on identifying predictors of speeding. Chapter 4 provides the methods and results of the focus groups. Chapter 5 provides our conclusions from this study. Chapter 6 provides a brief discussion of some methodological considerations relevant to our results and conclusions regarding the data collection and analyses. Seven appendices are provided in Volume III of this report. They are: o Appendix A: Description of Data Processing Activities; o Appendix B: Description of Data Validation Activities; o Appendix C: Regression Analyses with Socioeconomic Control Variables; o Appendix D: Factor Analysis; o Appendix E: Personal Inventory Questions; o Appendix F: Personal Inventory Responses; and o Appendix G: Draft Moderator Guide and Handouts for the Phase 2 Focus Groups.
4
Chapter 2: Methods
Introduction
This study employed an on-road, naturalistic approach to investigating drivers’ speed choices and speeding behavior in both Seattle, Washington (urban driving), and College Station, Texas (rural driving). GPS units were installed in the 164 participants’ vehicles for three to four weeks and participants were asked simply to engage in their normal driving activities. Location and vehicle speed were derived from the GPS data and the posted speed limits, referred to in this report as the posted speeds (PSs), on travelled roadways were obtained from State and county databases. These data were merged into a final data set to compare driving speeds to posted speeds. Below, we discuss the research methods used in this study, including: Experimental Design; Dependent Variables and Covariates; Participants; Apparatus and Test Instruments; Data Collection Procedures; Data Processing Activities; Definition of Speeding; and Selection of Speed Type/Threshold.
Experimental Design
The goal of this research project is to quantify the factors that influence drivers’ propensity to speed and to use this information to determine effective speeding countermeasures. The experimental design had to balance carefully the need to sample a representative population with the need to control variance by defining homogeneous populations. It was important to maintain as representative a sample as possible in the current study. In particular, this research represents a new, more detailed look at real-world high-speed driving observed on the road with the potential to provide more nuanced information about who is speeding, when they are speeding, and what the different types of speeding look like. For this reason, it was important to cast as wide a net as possible in order to catch as much of this new information as possible, even if it meant that we were left with higher variance as a result.
The experimental design was also tailored towards capturing segments of the population that are likely to engage in speeding behavior, and for which we hope to develop countermeasures. Previous studies have shown that age, gender, and geographical area influence the propensity to speed (De Waard et al., 1995; Rienstra & Rietveld, 1996; Yagil, 1998). The experimental design in the proposal reflected these influences, combined them in a study consisting of a 23 full factorial design with age, gender, and geographical area as between-subjects factors with 164 drivers, and with data from each driver collected over at least three weeks.
5
Based on a power analysis conducted during the planning phase of this study, a small range of driver ages was excluded from the experimental design. This was done because overall statistical power was determined to be low and restricting the age variable to certain ranges was a simple way to reduce within-group variability. Table 2 shows the age ranges used in the study: 18 to 25 years old for the younger group and 35 to 55 for the older group. This change provided more homogeneous groups, which should have correspondingly reduced the variance and increased the statistical power. This change also provided a clearer separation of the groups so that the young drivers (i.e., younger than 25)—if they have a greater propensity to speed—are not mixed with those 25 to 30 years old, a transition point in the age range to generally safer drivers. From our power analysis, differences between drivers “younger than 30” and “older than 30” would require a greater number of drivers to be discernible. As a result, we chose to focus our sampling on drivers 18 to 25 and 35 to 55.
Table 2. Study design and allocation of participants. Geography Urban (Seattle, WA) Rural (College Station, Texas) Gender Male Female Male Female Age 18-25 35-55 18-25 35-55 18-25 35-55 18-25 35-55 Subjects Ss 1-20 Ss 21-40 Ss 41-60 Ss 61-80 Ss 81-100 Ss 101-120 Ss 121-140 Ss 141-160
Dependent Variables and Covariates
In order to accommodate the range of research questions in this study, several different types of data were collected from participants. A summary of the types of dependent variables is presented in Table 3 below. Note that each of the data categories listed in the left-most column was comprised of multiple, more specific variables. More detailed information about each is provided in the relevant sections below or as separate appendices (i.e., for the personal inventory data).
Table 3. Data used in the analyses. Data Data Type Data Collection Method Demographic Between Ss Categorical Participant Screener & Vehicle variables Between Ss Categorical Pre-Drive Questionnaire Personality/risk taking/attitude Between Ss Qualitative Pre-Drive Questionnaire Driving behavior (speed-related measures) Within Ss Continuous On-Road GPS Driving situational factors (roadway class, Within Ss Categorical time, day, trip length, etc.) Trip situational factors (purpose, passengers, Within Ss Categorical Driving Logs weather, stress, congestion, lateness, etc.) Attitudes, beliefs, and intentions related to Between Ss Qualitative Post-Drive speeding Questionnaire
The primary dependent measures involve variables that quantified the participants’ travel speed relative to the posted speed. The travel speed was taken from the GPS device, while the posted
6
speed information was extracted from county and State GIS data sets (and validated with Google StreetView™ and on-road driving; See Appendix A for a detailed discussion of data processing activities and Appendix B for a detailed discussion of validation activities). Several other speed- related measures were also recorded and/or computed, such as median speed, and maximum acceleration/deceleration rates (see Table 4 below). In addition to speeding-related variables, other information about the driving situation (road type, trip duration, cumulative heading change, etc. and data-verification variables) were recorded during each driver trip.
Table 4. Driving-related variables recorded for each 30-second driving epoch/record. Type of Variable Elements of Variable Driving Average and median travel speed, cumulative heading change, posted speed, percentage of epoch spent in various speed bands. Speed Variability Standard deviation, minimum, maximum, standard deviation relative to exponentially weighted moving average. Descriptive Road name, road type, county, ZIP, trip ID, trip duration, spatial position. Variables and Flags for Checking Validity GPS signal, spatial-join error, missing GIS data, change tracking, etc.
Participants
A total of 164 participants fully completed the study requirements at both locations. Table 5 below shows the participant breakout by location, age group, and gender. Note that some participants have fewer days of driving data for a variety of reasons, including partial equipment failure, or personal activities limiting the number of their driving days during the study (e.g., vacation). Another noteworthy aspect is that the number of completed older male drivers in Texas (16 drivers) is lower than the objective for that condition (20 drivers). Although the data collection period was extended and participant compensation increased to recruit more of these drivers, interest among this group was so low that it was not possible to reach the data collection objectives. In order to compensate for this shortcoming to a limited extent, additional older male drivers were recruited in Seattle.
7
Table 5. Distribution of participant demographics. Urban – Seattle, WA Gender Male Female Total Age 18-25 35-55 18-25 35-55 Objective 20 20 20 20 80 Total Scheduled 35 35 27 30 127 Total Enrolled 22 25 22 22 91 Withdrew/Hardware Failure 1 0 1 1 3 Total Complete 21 25 21 21 88 Rural – College Station, Texas Gender Male Female Total Age 18-25 35-55 18-25 35-55 Objective 20 20 20 20 80 Total Scheduled 25 17 24 25 91 Total Enrolled 22 17 22 22 83 Withdrew/Hardware Failure 0 1 1 2 4 Total Complete 20 16 21 19 76 Grand Total Complete 41 41 42 40 164
Participant Recruitment
Participants were recruited using multiple approaches, and the specific approaches used varied based on location because of the greater difficulty of obtaining eligible participants in Texas than in Seattle. Both locations used newspaper and internet classified advertisements, in addition to posting flyers at local colleges, community centers, and libraries. These approaches were sufficient to reach urban/suburban drivers in Seattle who did most of their driving in built up areas. However, obtaining rural drivers in Texas was a more difficult task, and additional measures had to be taken at that location, such as setting up information tables at locations frequented by rural drivers, distributing post cards, placing ads on local radio stations, handing out business cards at hardware stores, and sending emails to clubs. These approaches were supplemented by an informational Web site. The Web site was advertised in all recruiting materials that contained detailed information about the overall study, participant eligibility requirements, and what participants could expect to do during the study. Potential participants were given a screening interview over the phone.
Participant Incentive Structure
Participant financial compensation contained two components:
1. Fifty dollars ($50) for the primary participation stipend (completion of questionnaires, most log entries, on-road data collection); and
2. Fifty dollars ($50) for the logbook completion incentive (which required approximately 80 to 90% or greater completion rates for logbook entries).
8
At the end of the study, it became increasingly difficult to recruit participants. This was particularly the case for older male drivers in Texas; on multiple occasions during recruiting interviews, these drivers indicated that total compensation levels were too low. In order to meet recruiting objectives, compensation was increased to $200 for each participant.
Apparatus and Test Instruments
This section describes three devices that were used to characterize personality and related factors, situational factors while driving, and driving behavior. These devices include:
1. Pre-drive and post-drive personal inventories;
2. Trip logs; and
3. GPS-based driving data.
Each of these devices is described below. The GPS-related information includes device selection, a description of the chosen device, and device preparation and installation.
Pre-Drive and Post-Drive Personal Inventory A personal inventory was developed to measure demographic, personality, attitudinal, and other individual factors associated with speeding behavior. The set of questions covered topics identified as having the greatest potential to account for variance in on-road speeding data. These included elements such as demographic factors, sensation seeking attitudes, risk-taking behaviors, and attitudes and beliefs about speeding. To the extent possible, we incorporated existing measurement scales that had already been validated (Sensation Seeking Scale, Driver Behavior Questionnaire (DBQ), etc.). Table 6 below shows the types of questions included in the personal inventory.
One of the key issues that was necessary to address in question set development was to minimize the potential for “contaminating” driver behavior if participants became aware that the primary objective of the study was to investigate speeding. In particular, because the questions covered several different aspects of speeding behavior, there was a real possibility that drivers might be alerted to the study objective of recording speeding behavior. The concern was that this could cause some drivers to change their actual speeding behavior to avoid being “caught breaking the law.” The simplest solution to this problem was to exclude most speeding-related questions from the pre-drive inventory, and instead move them to the post-drive inventory (see below). In this case, it would not matter to the on-road portions of this study if drivers guessed the study objective at that time, since the driving data already would have been collected by that point.
9
Table 6. Questions included in the personal inventory.
Start-Up Inventory Scope Number of Questions Comment (15-20 min) Demographic Questions Age, income, education, vehicle, 10 Questions were modeled from etc. the 2000 U.S. Census questions. Travel Behavior Trip planning, route selection, etc. 5 Questions developed for this study. Driver Behavior Driving behavior (inattention/”slips,” 28 Modified from Reason et al., Questionnaire (DBQ) errors, and violations). 1990. CARDS and DeJoy Risk Risky driving behaviors. 18 Redundant questions with DBQ Perception and across both risk question Questionnaires Combined sets were removed. Zuckerman’s ZKPQ Imp- Sensation seeking and impulsivity. 9 sensation seeking, 10 Shortened and updated version SS scale impulsivity questions of Sensation Seeking Scale.1 Close-out Inventory Scope Number of Questions Comment (15-20 min) Theory of Planned Attitudes, beliefs, social norms. 39 Questions from Elliott et al., Behavior (TPB) 2005. Self-Reported Driving Typical comfortable driving speeds 4-Seattle Part of the post-drive Speeds on different types of roads. 7-Texas questionnaire.
Trip Log
In order to capture additional information on situational factors during trips, participants were asked to complete daily trip logs as a component of their participation. The logs consisted of easy-to-complete checklists, designed to capture the situational factors that may influence speed choice, including trip time and purpose, number of passengers, weather conditions, distraction, fatigue, alcohol use, and if someone other than the participant drove the vehicle during an outing. An example of the trip log form for a single trip is shown below in Figure 2.
1 Information about the Zuckerman-Kuhlman Personality Questionnaire (ZKPQ) is available from www.zkpq.com/principaling.htm.
10
Trip Date: ______Time of Day: ______AM / PM No. of Passengers: ______
Who Drove: Self Other
Purpose of Trip (check all that apply): Weekday Commute Appointment Recreational Errands Other
Road / Weather Conditions (check all that apply): Dry/Sunny Fog Light Rain/Drizzle Heavy Rain/Hail Snow/Ice
Ongoing Conditions/Factors During your Trip (check all that apply): You were late Traffic was You were You were You were Consumed heavy stressed out fatigued generally alcohol in distracted last hour
Figure 2. Daily trip log data fields for a single trip.
Although the trip logs were simple to complete, doing so consistently required a relatively high degree of participant involvement. Accordingly, several steps were taken to encourage trip log completion, including:
• Communicating the trip log activities as a study priority to participants. This was effective because the study was described to participants as a “travel” study that involved collecting data about their route selection and factors that affect this, which made the trip logs highly relevant. Also, several of the logbook data fields were consistent with this explanation (trip purpose, congestion, levels, etc.). Accordingly, logbook completion was described as a primary objective of the study, with the understanding that they were expected to devote effort to this task as part of the participation requirements.
• Making log completion easy. Participants were provided with three different ways to complete the daily log information, including: (1) small booklets that participants could keep with them in their vehicles, (2) phoning in their entries, and (3) online completion. Also, a trip entry consisted of only a limited number of easy-to-complete check-box fields, with a few requiring written entries (e.g., time, date). In addition to these written logs, each participant was e-mailed a daily reminder with a link to the online version of the logbook.
• Providing financial incentive for consistently completing the trip logs. Participants were informed that they could receive a significant monetary bonus if they completed all or most of the trip log entries ($50, the same amount as the value for study participation).
Note that the trip log information was not used in the current analyses, although it is available for future analyses. The primary reasons why these data were not used are listed below. Despite our effort to maximize completion rates, there were a relatively high number of missing trip log entries for most participants (i.e., approximately 50% of trips identified from the GPS data did not have corresponding trip logs).
11
Similarly, a few drivers that had usable driving data had no trip log data at all, so we would have had to eliminate these drivers from any analyses. We decided that this cost was too high. Trip log entries were often the same across most trips, and the key situational factors (being late, stressed, fatigued, etc.) were rarely indicated. Part of this had to do with the unusually dry weather conditions in Seattle during the data collection period.
Given the high level of effort to include trip log data in the data set compared to the limited use we had for those data in the current analyses, we decided to focus on other data analysis activities instead.
Device Selection and Pilot Testing
This section provides an overview of the process used to select and test the GPS device. See the project work plan (Richard et al., 2008) for a detailed description of this process.
GPS device selection involved careful consideration of trade-offs between device capabilities, device cost, staff labor for installing/removing the devices, costs associated with any required modifications to the GPS unit packaging, and labor for turning raw GPS outputs into data that is appropriate for statistical analysis. To facilitate device selection, a set of selection criteria was developed that included GPS performance requirements, data content and storage requirements, and operations-related requirements (required packaging modifications, installation options, power management, etc.). Table 12 in the work plan provides a complete list of these criteria. An assessment of over a dozen current devices was performed, and two units were identified as candidate systems. The two units were evaluated in a series of rigorous, side-by-side tests to determine which device offered the best overall performance and addressed the evaluation criteria in the most workable fashion.
GPS Device
The GeoChron GPS data logger, manufactured by SparkFun, was chosen as the GPS device to be used in the project. This data logger continuously records GPS data to a removable Secure Digital (SD) flash memory card with a maximum storage capacity of 2 GB. The data are recorded to the SD card in human-readable form using standard National Marine Electronics Association (NMEA)2 sentences at a rate of one sample per second. The on-board GPS module is based on the SiRFIII chipset and is specified to exhibit positional accuracy of 5 m or better with Wide Area Augmentation System (WAAS)3 enabled and speed accuracy of 0.1 m/s (0.22 mph). The GPS module includes a built-in antenna as well as a connector for attaching an optional external antenna. The unit operates on a range of voltages (6 V to 14 V), and is designed
2 The GeoChron GPS module outputs the location, speed, heading, and other data using a standardized format published by the NMEA. The GeoChron supports a small subset (SiRF, 2005) of the sentences available in the NMEA 0183 standard; in our implementation, we recorded only the Recommended Minimum Communication (RMC) sentences specified in the standard.
3 WAAS (FAA, 2009), is a GPS error correction system that uses a network of satellites that transmit messages to precisely surveyed ground stations. These messages contain information that GPS receivers use to correct errors in the GPS signal and improve location accuracy and reliability.
12
to record a new data file each time power is applied to the power connector. Finally, the unit is packaged in an enclosure that was easy to modify and made it possible to develop multiple methods of mounting the unit in the vehicle. Figure 3 shows the GeoChron as it was received from the manufacturer.
SD Memory Card
Status LED Control Switches
Figure 3. GeoChron GPS data logger prior to modification.
A 2-GB SD flash memory card was used to capture the GPS data. This high level of memory capacity ensured that ample memory was available to capture all of the data for a participant, particularly for those vehicles in which the unit was continually powered for the duration of his/her trial.
Device Preparation and Installation
Although the GeoChron data logger was well suited for our application, some modifications to the unit were required in order to prevent participants from tampering with the unit, to manage power, and to affix the mounting hardware to the data logger. This section describes the modifications that were made to the GeoChron to accommodate these requirements.
Tamper Proofing
The GPS installation was designed to be completely self-contained and require no participant interaction or intervention. We also wanted to prevent participants from inadvertently turning off the unit or removing the SD memory card from the side of the unit. In order to achieve these goals, the GeoChron’s power and standby switches were removed from the unit and a custom mounting adapter bracket was fabricated from ABS plastic with tabs that covered the switch mounting holes and prevented access to the SD memory card.
13
Power Management
The GeoChron includes an internal rechargeable battery for use without an external power source. The duration of a participant trial, however, was expected to far exceed the life of a single battery. Therefore, it was necessary to power the unit using an external power source. To simplify installation and minimize installation time, a 7.5 V regulated DC power adapter that plugs into the vehicle’s cigarette lighter power receptacle was chosen to provide power to the GPS. The power adapter cable was replaced with a 10-foot cable in order to accommodate installations in larger vehicles, such as pickup trucks and sport utility vehicles (SUVs).
The GeoChron charges the battery whenever power is applied to its power receptacle. However, it was discovered during testing that if the battery is drained completely, the charging circuit limits current to the battery in order to optimize the charging current and prevent over-charging the battery. In consequence, the unit does not supply sufficient current to power the GPS module while charging a completely drained battery. Because the unit is constantly active, the battery will drain once the ignition is turned off and power is not available at the vehicle’s power receptacle, resulting in significant data loss while the battery recharges.
The data logger charging circuit was modified to overcome this challenge. The GeoChron’s internal battery was not needed because the unit was powered by the vehicle’s battery, so the internal battery was disconnected from the circuit board, and the charging circuit and power switch were bypassed using a jumper wire to the data logger’s voltage regulator circuit. The effect of this modification was that the GeoChron was always active and logging GPS data whenever power was applied at the vehicle’s cigarette-lighter power receptacle.
Mounting Hardware
One of the requirements for the GPS unit was that it require minimal time to install in a variety of vehicles. To accomplish this goal, mounting hardware was developed to accommodate two methods for mounting the GPS in the vehicle: (1) windshield mount and (2) dash mount. In order to allow a GPS unit to use either type of mount, a custom adapter bracket was fabricated from ABS plastic that allowed rapid installation and removal of a windshield-mounting bracket. The windshield mount had three large suction cups for gripping the windshield and was adjustable to allow the experimenters to level the GPS unit after mounting it on the windshield. The suction cups were easily replaceable in the event that repeated use should reduce effectiveness of grip. Figure 4 shows the GPS installation using the windshield mount.
14
Figure 4. GPS installation using the windshield mount option.
The GPS could also be installed by placing it directly on the dashboard (see Figure 5 below). A non-skid rubber pad was affixed to the bottom of each GPS unit to prevent the unit from sliding during hard braking, sharp turns, or other sudden maneuvers. To provide additional mechanical security, a tether was inserted into a slot that was milled in the enclosure and attached to the vehicle either with a suction cup affixed to the windshield or else by tying it through one of the heater vent louvers. Also, a black felt cover was attached to the top of the unit to minimize glare of reflections from the GPS and to reduce conspicuity of the device on the dashboard.
Figure 5. GPS installation using the dash mount option.
The GPS units were not as robust as expected in the summer heat, particularly in Texas where temperatures as high as 187.5 degrees F. were measured on the dashboard. Although the GPS electronics were not damaged, the plastic enclosures became severely deformed under these conditions. In order to solve this challenge, a third, antenna-based approach was developed. In Texas, an external antenna was mounted on the roof of the vehicle using a magnetic mount, and the cable was routed through the doorframe to the GPS logger. In Seattle, the antenna was mounted inside the vehicle, in either the front or the rear windshield. In both locations, the GPS logger was placed under the front passenger seat or under the dashboard to protect it from direct sunlight. Figure 6 illustrates the antenna installations on the roof and inside the vehicle.
15
Figure 6. GPS installations using exterior and interior antenna mount options.
In order to allow this type of installation, the GPS enclosure was modified by drilling a hole in the body and mounting a pass-through connector for an external antenna. In addition, the batteries inside the GPS unit were removed as a precaution to prevent them from swelling and leaking due to overheating. All units in Texas were modified, and the antenna mount option was the only installation used at that location. In Seattle, only eight units were modified, and the antenna mount option was installed only if the participant indicated that he or she parked in the sun all day or had no air conditioning. In addition, the antenna mount option was used for a few participants who were concerned about thieves breaking into the vehicle and wanted the installation to be as inconspicuous as possible. The small size and inconspicuous placement of the antenna in the windshield was an acceptable alternative for those participants.
Device Installation and Removal Procedures
GPS installation required either five or six steps, depending on whether the antenna mount option was chosen for installing the GPS. These steps are described as follows:
Step 1. Determine the appropriate GPS mounting option. This step was performed in collaboration with the participant. The experimenter described the various mounting options and gave the participant an opportunity to express any preferences or concerns about where to place the GPS or how to mount it in the vehicle. The GPS was mounted after agreement on an appropriate mounting strategy.
Step 2. Mount the GPS antenna (antenna mount installation only). The antenna typically was placed on the dashboard near the right A-pillar or in the center of the dashboard near the windshield. In a few vehicles, the antenna was placed in the rear window. The
16
antenna cable was routed by tucking it in the seam where the windshield met the dash and then through the gap between the dash and the right A-pillar. If necessary, the antenna was secured to the dash with a cable tie through the heater vent louver.
Step 3. Secure the data logger to the vehicle. Windshield-mounted units were attached to the windshield near the right A-pillar using three large suction cups on a swiveling mounting bracket. The windshield was cleaned with a damp cloth and the GPS data logger was mounted on the windshield. The GPS was then leveled laterally and longitudinally using a spirit level to ensure maximum signal strength.
In the dash-mounted installation, the GPS loggers were placed on the dashboard either near the right A-pillar or in the center of the dash over the center stack. The placement of the GPS logger depended on the flatness of the dashboard’s surface, location and orientation of appointments (clocks, coin wells, etc.), and participant preferences. The tether was then attached to the windshield with a suction cup or, alternatively, was tied to one of the heater vent louvers in order to prevent dislocation of the unit in the event of a sudden maneuver.
In the antenna-mounted installation, the GPS data loggers were placed under the front passenger seat to conceal them and to protect them from radiant heat from the sun. In one participant’s vehicle, the data logger was mounted under the dashboard and secured with cable ties to existing hardware.
Step 4. Mount the power adapter. The 7.5 V regulated DC power adapter was installed in the vehicle’s cigarette lighter receptacle to power the GPS data logger. Participants who regularly used their receptacle to power another device were given the option to install a “Y” adapter that provided an additional power outlet.
Part way through data collection, it was determined that the power to most participants’ GPS was being intermittently interrupted because of loose connections between the power adapter and the power receptacle. In order to reduce these interruptions, a small cardboard shim was wedged into the socket between the power adapter and the socket wall. This shim ensured a tight fit of the power adapter in the socket and greatly reduced transient power interruptions.
Step 5. Route and dress the cables. The cables from the power adapter and the antenna (for antenna mount installations) were routed through seams in the dashboard trim, under the passenger-side floor carpet, or under the dashboard in order to make the installation as inconspicuous as possible and to prevent passengers from tangling in the wires. Care was taken to make sure the power adapter and cables would not interfere with normal operation of the vehicle or use of cup holders or other features in the vehicle.
Step 6. Inspect the installation. After installation, the vehicle was inspected to make sure that the GPS was properly installed, the cables were hidden, and the power to the GPS was functioning.
17
Removal of the GPS data logger was accomplished by performing steps 2 through 5 in reverse order. After removal, the vehicle was inspected to make sure that all items related to the installation were removed from the vehicle and that the vehicle had sustained no damage. The memory card was then removed from the data logger, and the data were uploaded to a data repository for processing and warehousing.
Data Collection Procedures
There were three basic steps involved in collecting data from participants. These included participant enrollment and start-up activities, on-road data collection, and close-out and debriefing activities.
Figure 7 below describes the sequence of events encountered by each participant.
Initial Visit 21-day+ On-road Data Collection Period Close-out Visit
- Informed Consent On-road Data Collection - De-install
- Participant Training (Logs) DailyTrip Logs - Post-drive Question Set
- Equipment Install - Debriefing &Payment
Pre-drive Question Set If Taken Home to Complete
Figure 7. Overview of participant activities.
Enrollment and Start-up Activities
All participants enrolled in the study by visiting the research facility at their location. Participants were guided through a comprehensive consent process, which included discussion of potential risks and how the GPS device would be installed in their vehicle (only one prospective participant declined during the consent process). Following the consent process, the GPS device was installed in their vehicle. While this work was being conducted, participants were given detailed instructions on how to properly complete the trip logs in the log books and online. Finally, participants were given an opportunity to complete the pre-drive questionnaire while they waited for the installation to be completed. Apart from scheduling a close-out visit, answering any questions during the study, or providing reminders if participants were not completing online log entries, there was no other contact with them until the close-out visit.
Another important aspect of the study was to avoid presenting it to participants as a study focused primarily on speeding behavior. There were two important reasons for this. The first was so that participants who often drive at high speeds were not discouraged from signing up for the study under the belief that they would be penalized if they were recorded speeding. The second important reason was to avoid a situation in which participants who normally drive at high speeds curtail their normal speeding behavior out of concern for getting caught speeding. Our overall strategy for dealing with this issue was to frame the study as more of a general travel
18
behavior study, emphasizing our interest in learning about where, when, and on what types of roads participants travel on, in addition to the speed at which they travel. More specifically, the study’s speeding aspect was not withheld from participants, but rather it was presented as one of several factors we were interested in measuring. Necessary disclosures and protection of driver confidentiality were incorporated into the consent and debriefing process, and the overall data collection and analysis protocols. Moreover, drivers were informed fully about the precise objectives of the study during the debriefing process. Although all participants were given an opportunity to withdraw from the study during debriefing, none chose to do so, and no participants expressed concern about the speeding nature of the study to either the researchers or the study’s Institutional Review Board (IRB) representatives.
On-Road Data Collection
Participants were asked to drive as they normally would for the duration of the data collection. While data collection was targeted to last at least 21 days per participant, it extended longer for some participants due to their individual schedules; the average length of data collection was about 25 days per participant. During this time, participants’ primary responsibility was to complete trip logs after each trip and enter them online within one day. Trip log completion rates were tracked daily, and participants that fell behind by a few days were contacted to remind them about the importance of completing the trip logs, or to find out if there was any systematic reason for the lack of entries (e.g., vehicle in the repair shop or short vacation).
Data collection was conducted from May to December 2009 in Seattle, and from July to December 2009 in Texas.
Close-out and Debriefing Activities
Once the data collection period was complete, participants returned to the research facility to have the GPS equipment removed. During this visit, participants completed the final close-out personal inventory with speeding-related questions. After this, they were debriefed about the full study objectives and any questions that they had were answered. Finally, drivers were paid for their participation based on their trip log completion rate. Since it was only possible to estimate their completion rate at that time, only participants that clearly neglected this task (e.g., no trip logs for more than one continuous week) did not receive full compensation for this activity.
We also took this opportunity to inform participants about the focus groups to be conducted in Phase 2 and inquired about their interest in participating in those as well. Those who agreed (which was almost everyone who expected to be in the same area during the focus groups) were asked to provide consent for us to retain their contact information until recruiting for the focus groups could begin.
Data Processing Activities
The purpose of data processing was to clean and prepare the raw data into a final data set that was ready for analysis. These raw data came from four sources:
19
1. Personal Inventories, 2. Trip Log Data, 3. GPS Data, and 4. GIS Data.
These raw data required varying degrees of processing in order to clean and massage them into a form that was conducive to analysis.
Figure 8 is a flowchart that describes the flow of the data during processing and the relationships between various stages as the data were processed into the final data set.
Figure 8. Data processing flowchart.
The raw GPS data were cleaned and joined with the GIS maps, which were composited from a variety of sources, in order to match the driving data with the underlying roadways. The integrated GPS/GIS data were then partitioned into 30-second epochs, and a set of 135 variables were calculated that characterized each epoch. Variables included: road characteristics; trip duration; time on road; posted speed; heading range; difference between travel speed and posted speed; minimum, maximum, average, and median speed and acceleration/deceleration; variability of speed and acceleration; and time below numerous speed thresholds associated with
20
an individual epoch. Several flags also were included that were used to assess the validity of the data within the epoch.
In a separate process, the trip log data were manually matched with the GPS/GIS data in order to link the data with the trip logs. At various stages in the process, the data were validated to ensure the results were of high quality and reliable. The final data set included three databases: 1. An epoch database. 2. A trip log database. 3. An Excel spreadsheet with the personal inventory data.
The epoch database deserves some additional explanation. In short, the real-time, on-road driving data obtained from 164 participants resulted in a very large data set of driving speeds and geographical locations that were—at least within trips—essentially continuous. The only practical way to analyze this type of data set is to condense the driving data in a manner that captures the underlying behaviors of interest and ignores irrelevant driving situations. In addition to the sheer number of data points representing speed choice/behaviors, different drivers under various driving conditions will employ different strategies for selecting and maintaining a desired speed. In order to address the effects of these different strategies and to remove time periods when drivers had no opportunity to speed (i.e., time periods when the driver is accelerating from a stop or decelerating, or when drivers are consistently driving at a speed that is below a threshold that represents “free-flowing” traffic conditions), we parsed the trips into 30-second epochs.
The three databases (epoch database, trip log database, and a database for the personal inventory data) were linked by a common participant identification code. The final output of the post- processor was an Access database for each participant that contained the 135 fields of data for each epoch. All of the individual participant databases were merged into one large database for final analysis.
Appendix A provides details related to each step performed during data processing.
Data Validation Activities
A key requirement throughout all aspects of the project was that the data used in the data analyses must be of the highest possible quality in order to obtain reliable results and maintain high confidence in our conclusions. There were a number of data integrity challenges and checks used throughout our post-processing activities associated with: the GIS data, the GPS data, spatial joins, the development of epoch data sets, and validating the trip logs. Appendix B provides a detailed description of the data validation activities.
Final Data Filtering
While the primary activity in data validation was to fix as much “problematic” data as possible, the final stage of this process involved filtering out any “problematic” data that remained. During the final filtering process, driving epochs were successively filtered to remove (1) missing
21
information, (2) inaccurate information, and (3) driving epochs in which participants had no chance to speed. Further details about these filtering stages are provided in Table 7 and Table 8.
Note that prior to conducting the filtering stages listed above, all epochs containing data points in which the vehicle was not in motion (i.e., vehicle speed <= 3 mph) were removed from the data set. There were two primary reasons for excluding these epochs from the actual data set. The first was that the GPS device did not function accurately when stopped or at very low speeds (i.e. the position information and heading had an unusually high degree of error). Second, the number of these epochs was somewhat arbitrary. In particular, most of these epochs represented time when the vehicle was stopped (perhaps even parked and with no occupant) but the GPS device was still recording, which occurred in any vehicle that supplied continuous power to the device. For this reason, the elimination of these was not considered part of formal filtering (hence the “step 0” label).
The tables on the following pages show the filtering results for the Seattle and Texas data. The filtering steps followed the same basic approach at both locations, with the following exception: With the Seattle data, all epochs containing posted speed changes had to be removed because of problems with overlaid roads/ramps. This specific problem did not occur with the Texas data. In fact, because of how the GIS data set was built from different sources, it was necessary to permit some degree of “flipping” (i.e., accommodating a flip or change to the wrong road and back). Permitting some flipping does not compromise the data since the modal posted speed during an epoch is used as the posted speed value, and the modal speed is largely unaffected by short- duration flipping.
The filtering described above resulted in what we labeled the “free-flow” epochs, since epochs in which drivers were traveling at least 5 mph below the posted speed likely represent times when they are stuck in traffic, slowing to turn, or responding to a traffic control device. Although epochs are not used in the primary analyses, they were used to obtain an overall picture of the amount of free-flow and speeding driving data available for analysis, and to generate maps showing where free-flow driving occurred. Accordingly, filter stage 4 in Table 7 and Table 8 shows the filtering step used to extract “free-flow epochs.” In particular, epochs that had more than 15% of the driving time at speeds below the free-flow threshold (posted speed minus 5 mph) were removed. It should be noted that the data set used to conduct the descriptive and inferential analyses did not use this filtering stage. Specifically, all one-second records with free- flow and speeding time were extracted from the validated data set (post Stage 3) regardless of how much free-flow time was present in an epoch (i.e., these points were pulled from both free- flow and non-free-flow epochs). Essentially, epochs were used to eliminate “non-driving” or “invalid” data, but beyond that, we returned to one-second records for data analysis.
There was another stage of filtering conducted with the free-flow data set to isolate the epochs that were defined as speeding events. This filtering is described in the Results section as part of the discussion about the definition of speeding events.
22
Table 7. Seattle data filtering results. Filter Step Epochs at Start of Epochs at End of Filter Stage (Criteria for Removing Epochs) Step Step Stage 0: Remove non-driving data (vehicle not Vehicle is stopped or traveling less than 3mph for the duration of the epoch.* 255885 163842 moving). Stage 1: Remove missing or incomplete data. Posted speeds less than 15 mph. 163842 149809 More than 5 missing GPS points. 149809 149005 Epoch length less than 25 sec (out of 30). 149005 143539 Stage 2: Remove invalid or incorrect data. Posted speed changes during the epoch. 143539 120709 Max acceleration > 5 m/s2. 120709 120534 Max deceleration > 5 m/s2. 120534 120435 Absolute value of heading range > 180. 120435 120033 Stage 3: Remove data on unvalidated FlagRoadValRatio ≠ 1. 120033 76539 roadways. Stage 4: Remove data where there is no Traveling slower than posted speed minus 5 mph, more than 15% of the epoch. 76539 43363 chance to speed. Stage 5: Remove data that are not speeding Traveling <10 mph above the posted speed (defines Type 3 speeding 43363 4164 events. events).**
Traveling <20 mph above the posted speed (defines Type 4 speeding events). 43363 111 * This row is included to account for all of the data collected (e.g., when the GPS device had power), but these epochs are not considered to be analyzable data because the vehicle is stopped or almost stopped. ** Type 3 speeding events do not include Type 4 speeding events.
23
Table 8. Texas data filtering results. Filter Step Epochs at Start of Epochs at End of Filter Stage (Criteria for Removing Epochs) Step Step Stage 0: Remove non-driving data (vehicle not Vehicle is stopped or traveling less than 3mph for the duration of the epoch.* 180565 114894 moving). Stage 1: Remove missing or incomplete data. Posted speeds less than 15 mph. 114894 85332 More than 5 missing GPS points. 85332 85188 Epoch length less than 25 sec (out of 30). 85188 81554 Stage 2: Remove invalid or incorrect data. Posted speed changes more than once during the epoch. 81554 76616 Max acceleration > 5 m/s2. 76616 76568 Max deceleration > 5 m/s2. 76568 76546 Absolute value of heading range > 180. 76546 76450 Stage 3: Remove data on unvalidated FlagRoadValRatio ≠ 1. 76450 59138 roadways. Stage 4: Remove data where there is no Traveling slower than posted speed minus 5 mph, more than 15% of the epoch. 59138 28822 chance to speed. Stage 5: Remove data that are not speeding Traveling <10 mph above the posted speed (defines Type 3 speeding 28822 1114 events. events).** Traveling <20 mph above the posted speed (defines Type 4 speeding events). 28822 34 * This row is included to account for all of the data collected (e.g., when the GPS device had power), but these epochs are not considered to be analyzable data because the vehicle is stopped or almost stopped. ** Type 3 speeding events do not include Type 4 speeding events.
24
Definition of Speeding
One of the key challenges in a study focused on understanding drivers’ motivations for speeding is defining what actually constitutes speeding. There are differences between the legal definition and other aspects, such as safety and what drivers commonly view as acceptable speeds. Moreover, the inferential analysis conducted in this effort required that speeding be categorized into time in which drivers are considered to be “speeding” versus “not speeding.” In order to address these issues, it was first necessary to define these States based on the travel speeds recorded in this study.
A review of the research literature indicates that there is no single accepted definition of speeding, and a variety of approaches have been used (see Table 9 below). For example, many questionnaires that cover speeding ask about driving 10-20 mph above or greater than 20 mph above the posted speed (e.g., DeJoy, 1992), whereas others ask about driving faster than other drivers (e.g., passing most drivers). Other approaches have included defining speeding relative to law enforcement practices, which captures the psychological effects that avoiding a speeding ticket likely play in determining driving behavior (Book & Smigielski, 1999). Another common approach is to use posted speed plus a fixed amount (Warner & Aberg, 2006). This is a reasonable approach, since other studies suggest that posted speed is the most significant predictor of aggregate travel speed, although other factors, such as road design, are also important in many circumstances (Fitzpatrick, Carlson, Brewer & Wooldridge, 2000; Giles, 2004).
The reality is that there are multiple ways to define speeding, and each lends itself to particular research objectives, such as understanding why drivers do not comply with posted speed, or why they are willing to drive at “risky” speeds, etc. Drivers’ willingness to comply with posted speeds provides important information for developing countermeasures that rely on changing driver motivations and attitudes. Similarly, defining speeding relative to crash risk is also useful in that it helps identify the driver and environmental factors that are most likely related to safety, and which could be potentially addressed by a different set of countermeasures. Consequently, it is important to be mindful of the overall research objectives when selecting a definition of speeding.
One problem that most definitions of speeding encounter is how to set the specific threshold that divides driving into “speeding” and “not-speeding.” There are two key issues:
1. Whether or not driving constitutes speeding depends greatly on the particular driver and situation, and
2. Even if these driver and situational variables are known, there is not enough other empirical evidence to claim with any certainty that driving below a certain speed is safe/acceptable whereas driving above that speed is not.
Consequently, there is an unavoidable degree of arbitrariness in whatever approach is selected for defining speeding. However, in spite of this, it is still important to try to minimize this arbitrariness by avoiding certain approaches that rely solely on data-driven, post-hoc definitions
25
of speeding (e.g., to define speeding as the fastest 15% of driving). Although this approach can be used to ensure that there are a sufficient number of speeding episodes available for analysis, it loses the connection to the underlying driver behaviors and decisions because the threshold is arbitrary (i.e., wherever the 15% mark is).
Table 9 provides an overview of relevant approaches for defining speeding, as well as the relative advantages and disadvantages of each approach.
26
Table 9. Different definitions of speeding. Speed Rationale Value/Definition Advantages Disadvantages Definition Analytical Uses a fixed criterion relative to the posted speed. Posted speed + X%, or Simple to implement. Overly simplistic and misses factors Posted speed + X mph related to road type and overall speed level. Risk/ Reflects “driving too fast for conditions.” Ideally Specific thresholds that depend Connects speed behavior to Insufficient GIS data scope and Kinematic takes into account what speed is acceptably safe on data about driving primary safety objectives. quality to support reliable “risk” based on the immediate roadway/driving conditions (road class, posted classification. environment, weather, time-of-day, driver, and speed, time of day, weather, vehicle factors. etc.). Enforcement Some drivers may view this as the practical Posted speed + 11 mph, etc. − Some drivers implicitly take − Need information from local law definition of speeding—how fast can they drive into account likelihood of a enforcement to determine without getting a ticket. ticket in speed decisions. nominal values. − Should be simple to implement − Is probably an incomplete basis if we can obtain typical for understanding speeding enforcement values. behavior. Psychological Is directly based on what drivers consider to be Posted speed + 5 or 10 mph − Provides general insight into Requires additional information speeding. It can reflect factors that are specific to (the psychological space for which speeds drivers consider about speeding beliefs and individuals or groups, such as what seems to be speed likely consists of 5 mph to be speeding under certain attitudes from participants or safe or comfortable, which speeds will avoid increments). conditions. sample drivers that we did not enforcement, etc. − Can be used to provide driver- collect. specific speeding benchmarks. Behavioral Separate travel-speed bands that represent See Table 10 below. − Ranges are tied to plausible − “Clearly reckless” criterion is underlying behavioral intentions, including: driver behaviors and to safety somewhat arbitrary and may not 1) Intentionally trying to stay at or below the risk. generalize consistently across posted speed, but exceeding due to normal − Incorporates most useful different driving conditions. variation (control condition). aspects of other speeding − Complicates the analysis by 2) Deliberately exceeding the posted speed, but definitions. adding additional dependent unlikely traveling fast enough to be unsafe or variables. get a speeding ticket. 3) Traveling fast enough to increase safety risk and chances of getting a ticket, but not recklessly so. 4) Clearly reckless or excessive speeding.
27
For the analysis in the current project, we used a behavior-based approach to defining speeding. The basic rationale is that if drivers have control over their travel speed (not locked to traffic, etc.), then there is likely to be an underlying motivation or behavioral frame that is explicitly or implicitly governing their speed selection. There are perhaps several different types of speeding behaviors, different ways to conceptualize similar behaviors, and overlap between their corresponding ranges (both by definition and from natural speed variation). Note that there are other ways to define speed-related behaviors (inattentive speeding, versus deliberate speeding, etc.); however, the approach we selected appeared to provide the best way to differentiate categories based on travel speed with minimal overlap in expected speed ranges.
For the purpose of the analysis, we developed a framework and rationale for defining speeding, and then developed a way to implement the definition using the current data set. As a starting point, we focused on developing a behavioral framework that could segment driving into different categories reflecting different attitudes about speeding. In particular, four different behavior types were identified that correspond to different speed bands relative to the posted speed. These include: • Type 1: Driver is intentionally trying to stay below posted speed, which forms the control condition (ranges up to posted speed + natural speed variability). • Type 2: Driver is not concerned with or constrained by posted speed, but is trying to avoid getting a speeding ticket (ranges from posted speed to enforcement speed). • Type 3: Driver is unconcerned with posted speed or consciously willing to accept some risk (safety and speeding ticket) to drive faster (ranges from enforcement speed to what is clearly unsafe). • Type 4: Driver is traveling at a speed that most other drivers would clearly identify as “reckless” or too fast for conditions with a clear disregard for the posted speed.
These four types, translated into actual speed ranges (for a 60 mph posted speed), are shown in Figure 2 below (blue arrow bars), with hypothetical distributions showing speed variability around those ranges.
Type 1 Type 2 Type 3 Type 4
60 mph 70 mph 80 mph Posted Enforcement Speed Speed . Figure 9. Theoretical speed ranges.
28
Table 10 below provides further details about what these behavior types represent and how they would be used in the analyses.
Table 10. Speed behavior types. Speed Behavior Type 1 Type 2 Type 3 Type 4 Behavioral Driver is intentionally Driver is not concerned Driver is unconcerned Driver is traveling at a Intention trying to stay below with or constrained by with posted speed or speed that most other posted speed. posted speed, but is willing to accept some drivers would clearly trying to avoid a risk to drive faster. identify as reckless or too speeding ticket. fast for conditions. Numeric Upper bound is mostly Upper bound depends on Range is less certain Range is ultimately Range below posted perception of than lower groups artificial because it is speed+5 mph. Enforcement speed. because boundaries are defined by external more situation- criteria rather than dependent. specific driver behavioral intentions.* Driver No extra risk. Driver likely feels that Complicated relationship Safety risk is probably Perception of they are driving at a safe to risk: drivers may feel disregarded, or driver is Risk speed. safe even though they seeking out the “thrill” are not, or they may be associated with risk. accepting some risk. Relationship Should not involve Unlikely to be elevated Possibly elevated risk Clearly elevated risk. to Safety elevated risk (based on risk under most driving depending on conditions. roadway design). conditions. How It Could Non-speeder control Can be analyzed Speeding-related Speeding-related Be Used in condition/exposure. separately or combined independent variable. independent variable. the Data with non-speeder control Analysis condition since it likely does not involve “speeding of concern.” * A kinematic/risk-based approach would provide a more objective measure of this range.
A key practical problem we faced during the analysis was that the distributions around each band were unlikely to be as narrow or “tidy” as those shown in Figure 9, and that the initial data set looked more like a single undifferentiated normal distribution. Consequently, the true boundaries between behavior types were nearly impossible to discern with any degree of certainty. Nevertheless, this overall approach allowed us to maintain some degree of a priori constraints on how speeding events were defined and minimize opportunistic and arbitrary setting of the speeding criteria.
Selection of Speed Type/Threshold
Once the basic rationale for defining speeding was outlined, the next step was to develop a workable implementation using the current data set. As a starting point, we defined Type-3 speeding to be driving 10 to 20 mph above the posted speed and Type-4 speeding to be driving over 20 mph above the posted speed. These values were chosen primarily to provide round numbers to define the speeding categories, and give reasonable cutoffs for the underlying attitudes and motives that we are trying to capture. However, after examining the set of Type-4 speeding events, it was clear that there were too few events
29
to be statistically analyzed (particularly on the 70 mph roadways in Texas). Therefore, although they represent different behaviors, Type-3 and Type-4 speeding events were combined and are referred to simply as “speeding events” or “speeding time” hereafter. Accordingly, the threshold for categorizing driving as “speeding” in this report is 10 mph above the posted speed. 4
In both Seattle and Texas, a subset of the roadways with frequently traveled posted speeds were chosen for speeding analysis. Driving on roadways with different posted speeds was pooled to a general speed band. Driving on roadways with 30, 35, 55, and 60 mph posted speeds was combined to form two speed bands: 30-35 mph roadways and 55-60 mph roadways at both locations. (An additional 70 mph speed band was included in Texas, since this band contained the greatest amount of driving at that location.) This approach essentially involved pooling the data from roads with similar roadway environments, which allowed us to capture and examine more driving data. In general, the selection of two speed bands was important to examine certain posted speeds in isolation of others because of situation-specific factors systematically associated with posted speeds that likely affect driver behavior (e.g., 60 mph roads are more likely to be limited access highways). The drawback, however, is that it significantly increases the number of these separate analyses that have to be conducted.
Table 11 below shows the final thresholds used to define speeding-event epochs for posted speeds in each speed band. This table also provides a high-level view of the amount of free-flow and speeding driving available for analyses. Specifically, the numbers of driving epochs that qualify for each driving speed category are shown. This table also lists, in parenthesis, the free-flow and speeding time in terms of the number of seconds of driving in each of these categories (these one-second records ultimately comprised the data set used in the descriptive and inferential analyses). These individual seconds were taken from all validated epochs, not just the ones that met the free-flow epoch criterion (i.e., 85% of driving at free-flow speeds). More specifically, while many of the seconds represent the same driving as represented by the epochs, other seconds of driving were captured from non-free-flow epochs. For example, the Texas 70 mph speed band shows three eligible epochs (90 seconds worth); however, 121 seconds of driving were included in this speed band.
Defining “speeding” was a necessary first step of the data analyses. However, it still lacks a way to reflect exposure and begs the question, “amount of speeding relative to what?” Without a way to estimate overall exposure, directly comparing the amount of speeding across drivers, roadway types, geographic locations, etc., becomes problematic. To address this problem, a new variable was constructed from the data set—“free-flow time”—that could serve as the denominator to the “speeding time” numerator in our subsequent analyses. The free-flow time variable was carefully defined and constructed, with the objective of making sure that the measure reflected a driver’s opportunity to speed. For example, time spent warming up a vehicle was not included, nor was time spent idling at a stoplight. A less obvious consideration is that a drivers’ decisions about speed selection include many instances in which the vehicle’s speeds are constrained by traffic control devices or traffic congestion—instances in which drivers clearly have no opportunity to speed. If not accounted for, travel time that represents stopped time or slower driving could end up being overrepresented in terms of chronological time
4 Note that actual thresholds may have been even more conservative than our calculated thresholds. More specifically, during pilot testing, we found that test vehicle speedometers routinely displayed speeds as being around 2 mph faster than those actually recorded by the GPS. This is similar to results from other independent field tests (Markus, 2002), which showed a consistent 1-2 mph overestimation of travel speed by speedometers across 230 different vehicle makes. This suggests that a driver whose speed choice was based on the displayed speedometer speed may actually have been comfortable driving 1 or 2 mph faster than the speed recorded on the GPS.
30
relative to episodes in which drivers are able to select their desired speed, since drivers take longer to travel the same distance in the former case. Consequently, comparing the time that drivers spent speeding to the total amount of time they drove likely underestimates their speeding exposure by a substantial amount because their trip data is “diluted” by a large amount of inconsequential time when they were stopped or had no chance to speed.
The approach used to address this dilution of data was to remove from the data set all of the recorded time in which drivers clearly had no opportunity to speed. To generate the free-flow driving time variable, all driving that was 5 mph or more below the posted speed was removed (e.g., on a roadway with a posted speed of 35 mph, only driving time in which the vehicle was traveling at 30 mph or greater was retained). Relative to the complete data set, driving time was also filtered to remove driving that occurred on unvalidated (i.e., uncertain posted speed) roads, and due to other data-integrity issues, such as missing GPS data points.
Table 11. Final filtering thresholds used to define speeding-event epochs. The right side of the figure shows the number of 30-second epochs and seconds (in parentheses) retained. The Type 3 and Type 4 combined data were used in the analyses to represent speeding. Speeding Data Retained: Epochs (Seconds) Posted Speed Speed Band Threshold Types 3 & 4 Condition Free-flow Type 3 Type 4 (mph) Combined Seattle 30-35 mph 30 mph 40 mph 3052 (122672) 366 (4673) 2 (13) 368 (4686) Speed Band 35 mph 45 mph 4429 (172353) 414 (6130) 10 (156) 424 (6286) 55-60 mph 55 mph 65 mph 922 (29834) 57 (888) 1 (12) 58 (900) Speed Band 60 mph 70 mph 28586 (990439) 3327 (50329) 98 (1350) 3425 (51679) Texas 30-35 mph 30 mph 40 mph 1257 (62456) 105 (1759) 18 (324) 123 (2083) Speed Band 35 mph 45 mph 1864 (90601) 78 (914) 0 (0) 78 (914) 55-60 mph 55 mph 65 mph 1993 (75860) 319 (6101) 3 (101) 322 (6202) Speed Band 60 mph 70 mph 2588 (99919) 451 (11681) 10 (136) 461 (11817) 70 mph 70 mph 80 mph 15226 (525075) 161 (3840) 3 (121) 164 (3961) Speed Band
31
Chapter 3: On-Road Data Collection Results
Introduction
The following sections provide the initial results of the statistical analyses conducted in this study. The first section is the descriptive analyses, which provide a detailed description of the basic results. In an exploratory study such as this one, an in-depth understanding of the data is a critical first step for determining the most suitable approaches for answering questions about speeding behavior. The second section describes the inferential statistics conducted to examine, in more detail, the factors associated with speeding.
Descriptive Statistics
Overview
The descriptive statistics presented in the following sections provide results that cover a wide range of variables associated with speeding behavior. The scope is broad to the point that there is a risk that interesting findings are overwhelmed by the volume of information provided. This is somewhat unavoidable because of the complexity of the study, which involves separate locations, demographic groups, roadway factors (e.g., posted speed), and different types of measures (e.g., driving data and personal inventory data). To facilitate navigating the descriptive statistics, the rationale for our approach and an overview of the topics covered is provided below.
This more detailed examination of the data was a deliberate approach reflecting the nature of this study. In particular, this study represents a new way of looking at speeding behavior, and it has an exploratory nature to it. Essentially, going into the analyses, there were no established and proven approaches for investigating the type of speeding data obtained in this study. Also, in the absence of detailed hypotheses that we can use to focus the analyses, we think it is useful to describe the data on hand at a detailed level. This will make it easier to identify emerging patterns and trends that can be examined more closely in subsequent analyses.
A key challenge for presenting the descriptive statistics is that this study examined driving in a naturalistic context. More specifically, participants provided data at the time and location of their choosing, which means that we had no control over the driving conditions or other factors that influenced their opportunity to speed, and importantly, whether or not they chose to do so. This lack of control over driving conditions made the analyses vulnerable to systematic differences between groups and individuals in terms of key uncontrolled variables that affect speeding, which can lead to misleading patterns of results. For example, the older females recruited in Texas drove almost twice as many miles on 55 and 70 mph posted speed roads than the next closest demographic group. This driving pattern has repercussions for comparisons involving demographic groups, since the driving data used to make between-group comparisons is based on systematically different driving conditions.
Although driving epochs were used for filtering the data, the primary dependent measures were based on the free-flow driving that occurred at least 10 mph above the posted speed (i.e., Type 3 or Type 4 speeding). Two different dependent measures were used in the descriptive and inferential analyses.
32
These included (1) the percentage of trips that had any speeding, and (2) the proportion of free-flow driving that was speeding. These measures were computed using “free-flow time” and “speeding time” values calculated from the data. Driving data were grouped in speed bands (i.e., 30-35, 55-60, and 70 mph), and analyzed separately since these bands represent qualitatively different driving conditions. The key data elements are described in Table 12.
Table 12. Definitions of key terms used to calculate speeding measures. Term Definition/Usage Speed Band Identifies the posted speeds associated with a measure. The speed bands included in the analyses are: 30-35 mph (Seattle & Texas), 55-60 mph (Seattle & Texas), and 70 mph (Texas only). Free-Flow Time Number of seconds (typically in a single trip), for which a driver is traveling at speeds above the free-flow (FF Time) speed threshold (posted speed minus 5 mph). Free-flow time is a proxy for “opportunity to speed.” Free- flow time is separated out based on the speed band in which it occurred. Speeding Time Number of seconds (typically in a single trip) that the driver is traveling at 10 mph above the posted speed or faster (Type 3 and 4 speeding combined). Speeding time is separated out based on the speed band in which it occurred.
The descriptive results for Seattle and Texas are discussed separately, but sub-sections for each location share a common structure and cover the same basic data. The discussion of the analyses follows a specific progression that examines the basic elements of naturalistic driving and provides the initial understanding necessary for planning and conducting valid and applicable inferential analyses about the factors associated with speeding behavior. In particular, the descriptive analyses covered the following questions: What do we know about the drivers in the sample? (Personal Inventory Questions section) What can we say about their driving patterns overall? (All Driving Data section) What can we say about their opportunities to speed? (Free-Flow Time section) What can we say about their actual speeding behavior? (Speeding Time section)
The descriptive statistical sections that cover each type of data subset are described in more detail below.
Personal Inventory Questions
The first section of the descriptive statistics for each location provides a summary of participant responses to the personal inventory questions. The questions covering basic socio-demographic characteristics are detailed in separate charts, while the answers to other personal inventory questions are provided in a series of charts and tables that break down average participant responses by demographic category. The personal inventory was comprised of a series of question sets based on existing measurement instruments.
General Driving Patterns
This section covers the full, unfiltered driving data set. It is not the ideal data set for investigating specific aspects of driving behavior (and it is not used this way) because it contains episodes when vehicles were stopped and episodes with missing data (e.g., no posted speed information). However, it is the only data set that can provide information about driving patterns as a whole, in addition to basic trip
33
characteristics (duration, time of day, types of roads, etc.). For this type of high-level analysis, the imperfections in the data do not meaningfully affect the results, and the benefits gained by filtering out missing data, etc. do not offset losing useful information, such as trip start time or duration, that can only be computed with the full, unfiltered data set. If there was an option, however, we used dependent measures that were more robust regarding data issues when presenting findings (e.g., miles driven instead of time driven because it was unaffected by time when the vehicle was stopped).
Free-flow Time
This data subset represents the “cleaned” driving data that eliminates missing, inaccurate, or un-verified driving data (i.e., missing posted speed information) and, importantly, removes driving time in which participants clearly had no opportunity to speed. Free-flow time is used as the driving exposure measure (as opposed to epochs or miles driven). The summary of free-flow driving data provides basic information about driver speed choice patterns and information about some driving conditions, such as overall congestion levels.
Speeding Time
The speeding-time data subset comprised the key data for examining speeding behavior. This data subset contains all of the driving time (in seconds) that meets the criteria for speeding on each posted speed road (e.g., each second recorded above 70 mph on a 60 mph posted speed road). Two primary measures were used to quantify speeding behavior. The first was the proportion of a driver’s trips that had any speeding. This allowed us to examine the questions related to who is speeding. The second measure was the proportion of free-flow time on an individual trip during which the driver was speeding. This allowed us to examine the question related to who speeds the most. More details about how these measures are calculated and used are provided in the analysis of speeding time presented in later sections.
34
Seattle Data
This section discusses the descriptive statistics for the Seattle data set, including analyses of all personal inventory responses, general driving patterns, free-flow data, and speeding events.
Seattle – Demographic and Personal Inventory Data
This section provides more detailed information about who the drivers are in the study, and importantly, how the demographic groups generally differ in terms of their broader demographic characteristics, attitudes, behaviors, and beliefs, regarding driving. Drivers completed several personal inventories at the beginning and end of their participation in the study. These inventories were comprised of demographic and personality questions, generally organized by topic as follows: Demographic questions (based on US Census questions); General travel questions; Driver Behavior Questionnaire (DBQ); A combination of the DeJoy/Cards Risky Driving scales; The Sensation Seeking and Impulsivity scales from Zuckerman’s ZKPQ personality inventory; and Questions about attitudes, beliefs, and social norms regarding speeding based on the Theory of Planned Behavior (TPB).
The complete text of the personal inventories completed by participants is presented in Appendix C, which also includes the full set of response options for each question.
The following histograms describe the responses to the demographic questions. Note that the range of the Y-axis varies between figures in order to better show the differences between demographic groups.
35
Figure 10 below shows the distribution of household income levels for the Seattle area drivers. In general, older drivers have higher income levels than younger drivers.
35%
30%
25%
20% Older Females 15% Older Males Respondents Younger Females 10% Younger Males
5%
0% Under $15K - $30K - $45K - $60K - $75K - $90K + $15K <$30K <$45K <$60K <$75K <$90K Household Income
Figure 10. Histogram showing household income for Seattle drivers.
Figure 11 below shows the highest level of education obtained by drivers in each demographic group. All drivers obtained either a high school diploma or General Education/Equivalency Diploma (GED) and most drivers had at least an additional associate or bachelor’s degree.
60%
50%
40%
30% Older Females
Respondents Older Males 20% Younger Females Younger Males 10%
0% Did not High Associate Bachelors Masters Doctorate Other complete school/GED degree degree degree degree high school Highest Level of Education Obtained
Figure 11. Histogram showing highest level of education obtained for Seattle drivers.
36
Figure 12 below shows the marital status of the drivers in each demographic group. The younger drivers were mostly single, never married, while the older drivers were more distributed among categories. Most of the older drivers were married or had a partner.
100%
90%
80%
70%
60%
50% Older Females Older Males
Respondents 40% Younger Females 30% Younger Males 20%
10%
0% Single, Single, Single, Separated Married/partner never married divorced widow/er Marital Status
Figure 12. Histogram showing marital status for Seattle drivers.
Figure 13 shows the number of children that the Seattle area drivers had living with them. All of the younger drivers had no children. The older drivers were split among groups; most did not have children or had one or more children living with them most of the time.
120%
100%
80%
60% Older Females
Respondents Older Males 40% Younger Females Younger Males 20%
0% No One or more living One or more not living with you most of the with you most of the time time Status with Regard to Children
Figure 13. Histogram showing status with regard to children for Seattle drivers.
37
Figure 14 shows the distribution of self-reported weekly mileages for Seattle area drivers. Older Females tended to report higher mileages than the other groups.
60%
50%
40%
Older Females 30% Older Males
Respondents Younger Females 20% Younger Males
10%
0% Less than 50 50 - 100 100 - 200 200 - 300 300 + Self-reported Weekly Mileage
Figure 14. Histogram showing self-reported weekly mileage for Seattle drivers.
The personal inventory data from Seattle drivers is presented in a series of figures (Figure 15, Figure 16, Figure 17, and Figure 18) below. Given the large number of questions included in the personal inventory, only the three questions in each question set with the greatest range in responses (across all demographic groups) are shown. Accordingly, this subset provides an indication of the ways in which the demographic groups differ the most for a particular question set. Appendix D provides the average responses by demographic group for the full set of personal inventory questions. Each figure below shows the average response of each demographic group, with bars representing the 95% confidence interval.
38
Seattle – General Travel Questions
Question 11: How often do you check traffic conditions before you drive somewhere?
Older Females
Older Males
Younger Females
Younger Males
1 2 3 4 5 6 Never Hardly Occasionally Quite Frequently Nearly All Ever Often the Time
Question 12: During familiar trips (e.g., driving to work), how often do you change your travel route prior to departing to avoid congestion?
Older Females
Older Males
Younger Females
Younger Males
1 2 3 4 5 6 Never Hardly Occasionally Quite Frequently Nearly All Ever Often the Time
Question 16: How often do you leave early if you have to be somewhere at a specific time (e.g., work or an appointment)?
Older Females
Older Males
Younger Females
Younger Males
1 2 3 4 5 6 Never Hardly Occasionally Quite Frequently Nearly All Ever Often the Time
Figure 15. Results of general travel questions for Seattle drivers.
39
Seattle – Driver Behavior Questions
Question 20: How often do you ever attempt to pass someone that you hadn't noticed was trying to make a left turn?
Older Females
Older Males
Younger Females
Younger Males
1 2 3 4 5 6 Never Hardly Occasionally Quite Frequently Nearly All Ever Often the Time
Question 30: How often do you disregard the speed limits late at night or early in the morning?
Older Females
Older Males
Younger Females
Younger Males
1 2 3 4 5 6 NeverNever Hardly OccasionallyOccasionally QuiteQuite FrequentlyFrequently NearlyNearly All EverEver Often thethe TimeTime
Question 34: How often do you become angered by a certain type of driver and indicate your hostility in whatever way you can?
Older Females
Older Males
Younger Females
Younger Males
1 2 3 4 5 6 Never Hardly Occasionally Quite Frequently Nearly All Ever Often the Time
Figure 16. Results of driver behavior questions for Seattle drivers.
40
Seattle – Risk Questions
Question 45: In the past 3 months while driving, how often did you take risks while driving because it's fun, such as driving fast on curves or "getting air"?
Older Females
Older Males
Younger Females
Younger Males
1 2 3 4 5 6 Never Hardly Occasionally Quite Frequently Nearly All Ever Often the Time
Question 50: In the past 3 months while driving, how often did you accelerate when a traffic light turned yellow?
Older Females
Older Males
Younger Females
Younger Males
1 2 3 4 5 6 Never Hardly Occasionally Quite Frequently Nearly All Ever Often the Time
Question 59: In the past 3 months while driving, how often did you not make a full stop at a stop sign?
Older Females
Older Males
Younger Females
Younger Males
1 2 3 4 5 6 Never Hardly Occasionally Quite Frequently Nearly All Ever Often the Time
Figure 17. Results of risk questions for Seattle drivers.
41
Seattle – Speeding Beliefs, Attitudes, and Social Norms
Question 94: While driving in the next three months, how likely/unlikely is it that you would drive within or near the speed limit when you're late or in a rush?
Older Females
Older Males
Younger Females
Younger Males
1 2 3 4 5 6 7 Very Unlikely Somewhat Neither Somewhat Likely Very Unlikely Unlikely Likely Likely
Question 101: While driving in the next three months, how likely/unlikely is it that you would drive within or near the speed limit while driving on quiet roads in the day?
Older Females
Older Males
Younger Females
Younger Males
1 2 3 4 5 6 7 Very Unlikely Somewhat Neither Somewhat Likely Very Unlikely Unlikely Likely Likely
Question 102: While driving in the next three months, how likely/unlikely is it that you would drive within or near the speed limit while driving on quiet roads at night?
Older Females
Older Males
Younger Females
Younger Males
1 2 3 4 5 6 7 Very Unlikely Somewhat Neither Somewhat Likely Very Unlikely Unlikely Likely Likely
Figure 18. Results of speeding beliefs, attitudes, and social norms questions for Seattle drivers.
Seattle – Sensation Seeking and Impulsivity Subscale Scores
The 19 questions that were based on Zuckerman’s ZKPQ personal inventory can be summarized using two subscales: the Sensation Seeking subscale and the Impulsivity subscale. The Sensation Seeking subscale “describes the seeking of excitement, novel experiences, and the willingness to take risks for the sake of these types of experiences.” The Impulsivity subscale shows “a lack of planning and the tendency to act impulsively without thinking” (ZKPQ, n.d.). Figure 19 below shows the summary scores
42
for each subscale by demographic group. Younger drivers scored higher on the Sensation Seeking subscale, but all drivers scored comparably on the Impulsivity subscale. Higher scores correspond to higher Sensation Seeking and greater Impulsivity.
Zss
Older Females
Older Males
Younger Females
Younger Males
0.0 0.2 0.4 0.6 0.8 1.0
Zimp
Older Females
Older Males
Younger Females
Younger Males
0.0 0.2 0.4 0.6 0.8 1.0
Figure 19. Seattle – Sensation Seeking (Zss) and Impulsivity (Zimp) subscale scores by demographic group.
Seattle Data – All Driving
This section provides information about the overall driving patterns of study participants by demographic group. This includes information about when participants drove, for how long, and on what types of roads/speed bands. This information helps identify important trip-related factors that can be controlled for in the inferential analyses.
Seattle – Geographic extent of all driving
Figure 20 below provides a map of the lateral and longitudinal coordinates of the middle time point in each driving epoch (dots). The data points represent the location of each driver in 30-second increments. Driving epochs are presented here rather than time (i.e., one-second records) to reduce the number of points for a simpler and clearer map. Overall, these points indicate where the participants drove in the Seattle area. Most of the points are concentrated along the Seattle-Tacoma corridor, especially near the downtown core of Seattle, as indicated by the near-solid concentration of points in this area.
43
Figure 20. Map showing the geographic extent of all driving epochs in Seattle.
44
Seattle – Number of trips by duration and start time for all drivers combined
Table 13 below shows the number of trips that participants made as a function of the trip start time and the overall duration of the trip. Note that because the first and last half mile or first and last 90 seconds of travel (whichever was longer) was removed from each trip to protect participant privacy, 3 minutes was added back in to the trip duration to account for this deleted time. The table below shows that the majority of trips were between 5 and 30 minutes in duration and undertaken between the hours of 9 a.m. and 3 p.m.
Table 13. Number of trips by duration and start time for Seattle drivers. Time of Day Trip Duration (min) 12 a.m. -5 a.m. 5 a.m.-9 a.m. 9 a.m.-3 p.m. 3 p.m.-7 p.m. 7 p.m.-12 a.m. <5 12 82 319 194 162 10.7% 5-10 46 112 589 403 317 20.4% 10-20 55 243 867 639 463 31.6% 20-30 23 169 479 381 212 17.6% 30-40 13 138 194 242 99 9.6% 40-50 7 61 117 130 26 4.7% 50-60 5 18 54 58 20 2.2% >60 2 25 82 100 22 3.2% 2.3% 11.8% 37.6% 29.9% 18.4%
Seattle – Number of trips by duration and start time for each demographic group
Table 14 below shows the number of trips that participants in each demographic category made based on the trip start time and the overall duration of the trip. Again, 3 minutes were added to the duration of each trip to account for the time removed for participant privacy. The distributions of trips are similar across demographic categories. The most notable difference is that Older Females had a greater concentration of trips in the 9 a.m. to 3 p.m. time band than other groups. Additionally, younger drivers recorded more nighttime trips, between 7 p.m. and 12 a.m., than the older drivers.
45
Table 14. Average number of trips by duration and start time for each demographic group in Seattle. Older Females Time of Day Trip Duration (min) 12 a.m.-5 a.m. 5 a.m.-9 a.m. 9 a.m.-3 p.m. 3 p.m.-7 p.m. 7 p.m.-12 a.m. <5 1 11 76 49 26 9.4% 5-10 10 44 175 95 51 21.6% 10-20 4 47 268 183 59 32.3% 20-30 1 25 150 91 35 17.4% 30-40 1 40 66 45 23 10.1% 40-50 0 20 37 18 3 4.5% 50-60 0 5 17 11 2 2.0% >60 1 2 24 20 3 2.9% 1.0% 11.2% 46.8% 29.4% 11.6% Older Males Time of Day Trip Duration (min) 12 a.m.-5 a.m. 5 a.m.-9 a.m. 9 a.m.-3 p.m. 3 p.m.-7 p.m. 7 p.m.-12 a.m. <5 1 16 82 37 34 9.2% 5-10 4 29 140 104 66 18.6% 10-20 6 98 195 149 98 29.7% 20-30 2 78 113 104 37 18.2% 30-40 4 45 54 75 26 11.1% 40-50 1 25 42 52 12 7.2% 50-60 0 4 16 18 5 2.3% >60 0 16 15 36 1 3.7% 1.0% 16.9% 35.7% 31.3% 15.2% Younger Females Time of Day Trip Duration (min) 12 a.m.-5 a.m. 5 a.m.-9 a.m. 9 a.m.-3 p.m. 3 p.m.-7 p.m. 7 p.m.-12 a.m. <5 4 47 83 43 47 12.2% 5-10 16 22 146 103 82 20.1% 10-20 24 63 186 144 164 31.7% 20-30 11 39 112 120 72 19.3% 30-40 5 40 43 72 29 10.3% 40-50 2 9 16 30 4 3.3% 50-60 1 3 8 8 3 1.3% >60 1 2 14 12 3 1.7% 3.5% 12.3% 33.2% 29.0% 22.0% Younger Males Time of Day Trip Duration (min) 12 a.m.-5 a.m. 5 a.m.-9 a.m. 9 a.m.-3 p.m. 3 p.m.-7 p.m. 7 p.m.-12 a.m. <5 6 8 78 65 55 12.0% 5-10 16 17 128 101 118 21.5% 10-20 21 35 218 163 142 32.7% 20-30 9 27 104 66 68 15.5% 30-40 3 13 31 50 21 6.7% 40-50 4 7 22 30 7 4.0% 50-60 4 6 13 21 10 3.1% >60 0 5 29 32 15 4.6% 3.6% 6.7% 35.2% 29.9% 24.7%
46
Seattle – Number of trips by day of week for all drivers combined
Figure 21 below shows the total number of trips taken on each day of the week by all Seattle drivers combined. The day of the week is based on the trip start time. The number of trips taken is generally similar across weekdays, with many fewer trips taken on weekends. The number of weekend trips is likely smaller because participants tend to drive less on these days. However, another contributing factor is that most drivers were enrolled in the study for fewer weekend days than individual weekdays overall. Participant enrollments were not performed on the weekends, which reduced the opportunity to drive on one Saturday and one Sunday during the enrollment period for each participant.
1200
1000
800
600 Trips
400
200
0 Monday Tuesday Wednesday Thursday Friday Saturday Sunday
Day of Week
Figure 21. Histogram showing the total number of trips taken on each day of the week for all Seattle drivers combined.
Seattle – Average of trips by day of week for each demographic group
Figure 22 below shows the average number of trips taken on each day of the week (based on trip start time) for each demographic group. For most drivers, this represents two to three instances of each day of the week (e.g., 2-3 Mondays) depending on which days the start-up and close-out visits occurred and on the total duration of enrollment. Note that the average value is used in this table to compensate for the different number of participants in each demographic group. The differences across demographic groups appear to be minor, with the possible exception of Younger Males who drove more on Saturdays than the other groups.
47
16
14
12
10
Older Females 8 Trips Older Males Younger Females 6 Younger Males
4
2
0 Monday Tuesday Wednesday Thursday Friday Saturday Sunday Day of Week
Figure 22. Histogram showing the average number of trips taken on each day of the week for each demographic group in Seattle.
Seattle – Total miles driven by posted speed
Figure 23 below shows the total number of miles driven by all participants on roadways with the indicated posted speeds. The largest proportion of driving occurred on roads with either 35 or 60 mph posted speeds. This distribution of driving across posted speeds served as the basis for selecting these two posted speeds as the focus of our analyses. For the analyses, these speed bands were each combined with a neighboring posted speed, as denoted by similarly colored triangles in the graph below. The reasons for combining posted speeds were: (1) to increase the number of seconds of data in each speed band (particularly in Texas), (2) to enable the examination of the same posted speeds in both locations, and (3) because the road environments/driving conditions represented by the road types that were combined are similar.
48
35000
30000
25000
20000 Miles 15000
10000
5000
0 20 25 30 35 40 45 50 55 60 65 70
Posted Speed
Figure 23. Histogram showing the total number of miles driven by all participants on Seattle-area roads with different posted speeds (in mph). The colored triangles denote the posted speeds selected for analysis in the 30-35 mph speed band (purple) and the 55-60 mph speed band (green).
Seattle – Average miles driven on 30-35 and 55-60 mph posted speed roads for each demographic group
Figure 24 below shows the average number of miles driven on 30-35 and 55-60 mph posted speed roads by drivers in each demographic group. Note that the average value is used in this table to compensate for the different number of participants in each demographic group. Overall, there appear to be no systematic differences across demographic groups.
450
400
350
300
250 Older Females
Miles Older Males 200 Younger Females 150 Younger Males
100
50
0 30-35 55-60 Posted Speed
Figure 24. Histogram showing the average number of miles driven on 30-35 and 55-60 mph roads in Seattle by demographic group.
49
Seattle – Summary of trip-level data by demographic group
Table 15 describes some of the basic measures used to depict driving patterns per demographic group. This table shows that the trip-level characteristics were relatively consistent across drivers as measured by the average number of trips per day, average trip duration, and average trip distance. However, when looking at driving time, it appears that older participants drive more than younger participants do during weekdays, while Younger Males drive the most on weeknights and weekends.
Table 15. Key trip-level characteristics for drivers in Seattle. Older Older Younger Younger All Females Males Females Males Number of Participants (N) 21 25 21 21 88 Avg Days of Driving (unique days with trips) 21.2 20.3 22.4 20.4 21.0 Avg Trips per day 3.9 3.6 3.9 4.2 3.9 Avg Trip Duration (minutes) 20.6 21.9 19.1 22.0 21.0 Avg Trip Distance (miles) 9.1 10.0 8.6 9.4 9.3 Avg Driving Time Weekdays (hrs) 5 a.m.-7 p.m. 16.8 16.1 14.1 13.9 15.4 Avg Driving Time Weeknights (hrs) 7 p.m.-5 a.m. 2.1 2.4 3.8 5.0 3.3 Avg Driving Time Weekend (hrs) 12 a.m.-12 p.m. 4.7 4.3 4.7 6.6 5.0
In summary, there are some differences between groups with regard to the variables examined in this section, though the differences are small. However, if these differences systematically affect the opportunity to speed or a driver’s propensity to speed, they could potentially influence the overall findings. Some of these variables, therefore, were used as control variables in the inferential analyses.
Seattle Data – Free-Flow Driving
Seattle – “Free-flow” driving epochs
The data described in the previous sections were based on the full data set. One limitation of this data set with regard to examining driver speeding is that it includes many instances in which drivers’ speeds are constrained by traffic control devices or traffic congestion—instances in which they clearly have no opportunity to speed. Moreover, travel that represents slower driving or stopped time ends up being overrepresented in terms of chronological time relative to episodes in which drivers are able to select their desired speed, since drivers take longer to travel the same distance in the former case. Consequently, comparing the time that drivers spent speeding to the total amount of time they drove likely underestimates their speeding exposure by a substantial amount because their trip data is “diluted” by a large amount of inconsequential time when they were stopped or had no chance to speed.
One approach for addressing this dilution of data is to remove all of the recorded time (seconds) in which drivers clearly had no opportunity to speed. We did this in two different ways. For most of the descriptive and inferential analyses, we filtered individual one-second records of driving based on whether or not the vehicle speed was above the free-flow speed criterion. The other way we did this was to filter by epochs (this was necessary for making the GIS maps of free-flow driving). However, since epochs contained up to 30 one-second records--not all of which may have been at free-flow speeds--we
50
added a time-criterion for the epoch filtering. In particular, we only retained epochs for which driving was done within 5 mph of posted speed at least 85% of the time (e.g., traveling at 30 mph or greater on a 35 mph road for at least 85% of the time in an epoch). Table 16 below shows how many epochs were retained on 30-35 and 55-60 mph roads for each demographic category after this filtering step. Note that relative to the complete data set, free-flow epochs were also filtered to remove driving that occurred on unvalidated roads and due to other data-integrity issues, such as missing data points.
There is an important caveat with regard to using free-flow time as a measure of opportunity to speed. The key problem is that it can be a highly imprecise measure under certain conditions (i.e., light traffic). In particular, because we could not collect data about a driver’s immediate traffic environment, it is impossible to tell if a driver traveling at the posted speed (which we count as free-flow driving) is doing so deliberately or if they would rather be speeding but are (for example) trapped behind another vehicle that they cannot pass. A detailed description of how free-flow time was calculated for this study is provided on pages 58-59, below.
Table 16 shows how much of the validated driving data was done under free-flow driving conditions. In particular, just over half of the epochs on both 30-35 mph and 55-60 mph roads represented free-flow driving. Overall, there appeared to be little difference in these proportions across demographic groups.
Table 16. Comparison of the number of Seattle driving epochs in each demographic category before and after filtering to retain only free-flow driving epochs on 30-35 and 55-60 mph roads. 30-35 mph Posted Speed 55-60 mph Posted Speed Demographic Group Driving Epochs* Free-flow Percent Retained Driving Epochs* Free-flow Percent Retained Older Females 2776 1661 59.8% 11568 6347 54.9% Older Males 3448 2068 60.0% 15864 8630 54.4% Younger Females 3231 1975 61.1% 14325 7998 55.8% Younger Males 3006 1777 59.1% 11278 6533 57.9% Overall 12461 7481 60.0% 53035 29508 55.6% *Driving epochs include filtering to remove stationary vehicle data, missing/incomplete data, invalid or incorrect data, and data from unvalidated roadways (through Stage 3 of filtering as described in the Methods section).
Note that the percentage values in the table above are slightly higher than those found in Table 7 (on page 23) for the overall Seattle Stage 4 data filtering results. This is because the data obtained for driving on roads with 30-35 mph and 55-60 mph posted speeds was more reliable than the data obtained for driving on roads with other posted speeds. In other words, the amount of unreliable data from roads with posted speeds that were not selected was disproportionately high (due to missing data, invalid data, no chance to speed, or unverified posted speeds).
Seattle – Geographic extent of filtered free-flow epochs
Epochs can also show where in the road network the free-flow driving occurred. Figure 25 below provides a map showing two different types of epochs. The first represents all driving epochs (green dots), which are also the same set of data points as shown in the map in Figure 20 (page 44) above. The second set (blue dots) shows all of the epochs that remained after filtering for data quality/validity and for “opportunity to speed,” as described in the Methods section on epoch filtering. Essentially, the blue
51
dots indicate which of the total driving epochs (green dots) were included in the speeding analyses. Note that the green dots include all of the posted speeds, whereas the free-flow epochs shown (blue dots) represent only driving from 30-35 and 55-60 mph posted-speed roads.
As seen in Figure 25, the free-flow driving spans most of the overall driving area and it covers all of the major types of driving in the Seattle area (urban, suburban, freeway, and outlying areas with more open roads, or hillier terrain). Another notable pattern in the results is that on several roads the blue dots appear intermittently with green dots. This indicates instances of free-flow driving interspersed with slow driving, which raises questions about whether drivers actually had an opportunity to speed in these cases, even though they were traveling at free-flow speeds.
52
Figure 25. Map showing the geographic extent of the Seattle free-flow epochs.
53
Seattle – Percentage of free-flow driving time spent in different speed bands relative to the posted speed. Free-flow driving time can provide a comprehensive view of driver speed choice across all of the driving in the study. In particular, Table 17 shows the distribution of speeds relative to the posted speed across all of the Seattle drivers in a demographic group--shown in five-mile-per-hour bands (e.g., “PS+5” represents driving at speeds between 40-44.99 mph on a 35 mph road). The bold line between the “PS-5” (mph) and “PS” categories indicates the posted speed. Table 17 indicates that the majority of driving occurred within ±5 mph of the posted speed. Speeding in the current report was defined as driving that was 10 mph or more above the posted speed (categories shown in red text in the table below). In this case, the speeding that did occur was predominately between 10 and 15 mph above the posted speed. Almost no speeding occurred at speeds faster than 20 mph above the posted speed. In terms of this overall characterization of driving speed, there are some differences between demographic groups, but these differences are small.
Table 17. Percentage of free-flow driving time in Seattle spent in different speed bands relative to the posted speed (PS) for each demographic group on 30-35 and 55-60 mph roads. 30-35 mph Roads Travel Speed (mph)
PS-5 PS PS+5 PS+10 PS+15 PS+20 PS+25 Older Females 38.8% 43.7% 14.8% 2.5% 0.2% 0.0% 0.0% Older Males 34.0% 43.5% 17.8% 3.7% 0.8% 0.1% 0.0% Younger Females 33.2% 42.3% 20.7% 3.5% 0.2% 0.0% 0.0% Younger Males 31.9% 44.9% 19.6% 3.1% 0.4% 0.0% 0.0% Overall 34.4% 43.6% 18.3% 3.2% 0.4% 0.0% 0.0%
55-60 mph Roads PS -5 PS PS+5 PS+10 PS+15 PS+20 PS+25 Older Females 34.8% 42.4% 19.9% 2.6% 0.3% 0.0% 0.0% Older Males 31.6% 46.2% 17.6% 3.6% 0.7% 0.1% 0.0% Younger Females 25.1% 39.9% 28.2% 5.9% 0.7% 0.1% 0.0% Younger Males 24.1% 42.4% 27.3% 5.1% 0.8% 0.2% 0.0% Overall 28.9% 42.8% 23.1% 4.4% 0.7% 0.1% 0.0%
Seattle Data – Speeding Time A detailed analysis of the speeding time on each trip was used to examine patterns in driver speeding behavior. In particular, we focused on two primary measures to quantify speeding behavior. The first was the proportion of a driver’s trips that had any speeding. This allowed us to examine the questions related to who is speeding. The second measure was the proportion of free-flow time on an individual trip that was speeding. This allowed us to examine the question related to who speeds the most. More detailed discussion of each measure and how each was calculated is provided below.
54
Seattle – Percent of free-flow trips with speeding time by demographic group
This measure represents the percentage of trips during which the driver took advantage of the opportunity to speed, or the likelihood that speeding occurred on any particular trip for a driver. The percentage of trips with speeding time provides a high-level measure of the propensity to speed on a given trip, in addition to providing a way to identify which drivers engage in any speeding. Note that this measure only counts trips that contain free-flow driving in the specified speed band, and the speeding time is also specific to that speed band. The following definition shows how the percentage was calculated: